Meaning in Mathematics

Meaning in Mathematics
This page intentionally left blank

Meaning in Mathematics
Edited by
John Polkinghorne
1
3
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© Oxford University Press 2011
The moral rights of the authors have been asserted
Database right Oxford University Press (maker)
First published 2011
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Library of Congress Control Number: 2011920646
Typeset by SPI Publisher Services, Pondicherry, India
Printed in Great Britain
on acid-free paper by
Clays Ltd, St Ives plc
ISBN 978–0–19–960505–7
1 3 5 7 9 10 8 6 4 2
In grateful memory of Peter Lipton, scholar and friend.
Contents
List of contributors ix
Introduction 1
John Polkinghorne
1 Is mathematics discovered or invented? 3

Timothy Gowers
Comment 13
Gideon Rosen
2 Exploring the mathematical library of Babel 17

Marcus du Sautoy
Comment 26
Mark Steiner
3 Mathematical reality 27
John Polkinghorne
Comment 35
Mary Leng
Reply 39
John Polkinghorne
4 Mathematics, the mind, and the physical world 41

Roger Penrose
Comment 46
Michael Detlefsen
viii C O N T E N T S
5 Mathematical understanding 49
Peter Lipton
Addendum 55
Stewart Shapiro
6 Creation and discovery in mathematics 61

Mary Leng
Comment 70
Michael Detlefsen
7 Discovery, invention and realism: Gödel and others

on the reality of concepts 73
Michael Detlefsen
Comment 95
John Polkinghorne
8 Mathematics and objectivity 97

Stewart Shapiro
Comment 109
Gideon Rosen
Reply 112
Stewart Shapiro
9 The reality of mathematical objects 113

Gideon Rosen
Comment 132
Timothy Gowers
10 Getting more out of mathematics than what we put in 135

Mark Steiner
Comment 144
Marcus du Sautoy
References 147
Index 153
List of contributors
Editor: John Charlton Polkinghorne, KBE, FRS, the former president of

Queens’ College, Cambridge, and the winner of the 2002 Templeton Prize, has
been a leading figure in the dialogue of science and religion for more than two
decades. He resigned his professorship of mathematical physics at Cambridge
University to take up a new vocation in mid-life and was ordained a priest in the
Church of England in 1982. A fellow of the Royal Society, he was knighted
by Queen Elizabeth II in 1997. In addition to an extensive body of writing
on theoretical elementary particle physics, including Quantum Theory: A Very
Short Introduction (2002), he is the editor or co-editor of four books, the co-
author (with Michael Welker) of Faith in the Living God: A Dialogue (2001),
and the author of nineteen other books on the interrelationship of science
and theology, including Belief in God in an Age of Science (1998), a volume
composed of his Terry Lectures at Yale University, Science and Theology
(1998), Faith, Science and Understanding (2000), Traffic in Truth: Exchanges
between Theology and Science (2001), The God of Hope and the End of the
World (2002), Living with Hope (2003), Science and the Trinity: The Christian
Encounter with Reality (2004), Exploring Reality: The Intertwining of Science
and Religion (2005), Quantum Physics and Theology: An Unexpected Kinship
(2007), From Physicist to Priest (2007), Theology in the Context of Science
(2008), and Questions of Truth: Fifty-one Responses to Questions about God,
Science and Belief (2008) with Nicholas Beale.
Michael Detlefsen is McMahon-Hank Professor of Philosophy at the Uni-

versity of Notre Dame and Distinguished Invited Professor of Philosophy at
both the University of Paris 7-Diderot and the University of Nancy 2. He has
held a senior chaire d’excellence of the ANR in France since 2007. His chief
scholarly interests are in the history and philosophy of mathematics and logic.
His current projects include a book on Gödel’s incompleteness theorems with
Timothy McCarthy and various other projects concerning ideals of proof in
mathematics.
Marcus du Sautoy is professor of mathematics and Simonyi Professor for
the Public Understanding of Science at Oxford University, where he is a
fellow of New College. His academic work mainly concerns group theory and
x LIST OF CONTRIBUTORS
number theory, and he is widely known for popularizing mathematics. He was
awarded the Berwick Prize of the London Mathematical Society in 2001 and
the Faraday Prize by the Royal Society in 2009. He has presented numerous
series for BBC TV and radio and is the author of three books, The Music of
the Primes (2003), Finding Moonshine (2007), and The Num8er My5teries: A
Mathematical Odyssey through Everyday Life (2010), for general audiences.
Timothy Gowers, FRS, is Rouse Ball Professor of Mathematics at Cambridge
University and a fellow of Trinity College, Cambridge. He received a Fields
Medal in 1998 for his research connecting the fields of functional analysis
and combinatorics. Earlier, he was awarded the Junior Whitehead Prize by
the London Mathematical Society and the European Mathematical Society
Prize. A fellow of the Royal Society, he is the author of Mathematics: A Very
Short Introduction (2002) and the main editor of The Princeton Companion
to Mathematics (2008). Launched in 2009, his Polymath Project uses the
comment functionality of his blog to produce mathematics collaboratively.
Mary Leng is a lecturer in philosophy at the University of Liverpool. Her
research focus is the philosophy of mathematics, with particular reference to
issues raised by the applicability of mathematics in empirical science. Dr. Leng
has been a visiting fellow in the Department of Logic and Philosophy of
Science at the University of California at Irvine, and after a postdoctoral
fellowship in the humanities at the University of Toronto, she held a research
fellowship at St. John’s College, Cambridge, for four years, as well as a
visiting junior fellowship at the Peter Wall Institute for Advanced Studies at the
University of British Columbia. She is the co-editor (with Alexander Paseau
and Michael Potter) of Mathematical Knowledge (2007), and the author of
Mathematics and Reality (2010), both of which were published by Oxford
University Press.
Peter Lipton was the Hans Rausing Professor and chair of the History and
Philosophy of Science at Cambridge University until his death in 2008. He
also had been a fellow of King’s College, Cambridge. Much of his work con-
cerned explication and inference in science, but his interests extended broadly
across many of the major areas of philosophy. A fellow of the Academy of
Medical Sciences, he had been consulting editor of Studies in the History and
Philosophy of Science, the editor of Theory, Evidence and Explanation (1995),
and editor or co-editor of three special issues of Studies in the History and
Philosophy of Science. He was the author of Inference to the Best Explanation
(1991 and 2004).
Roger Penrose, OM, FRS, the Rouse Ball Professor of Mathematics Emeritus
at Oxford University and an emeritus fellow of Wadham College, Oxford,
is widely acclaimed for his original and broad-based work in mathematical
physics, particularly his contributions to general relativity theory, the foun-
dations of quantum theory, and cosmology. He also has written on the link
between fundamental physics and human consciousness. A fellow of the Royal
LIST OF CONTRIBUTORS xi
Society, a foreign associate of the National Academy of Sciences, and a fellow
of the European Academy of Sciences, Professor Penrose was knighted for his
services to science by Queen Elizabeth II in 1994 and awarded Britain’s Order
of Merit in 2000. He is the author or co-author of ten books, including The
Emperor’s New Mind: On Computers, Minds, and the Laws of Physics (1989),
winner of the 1990 Science Book Prize, Shadows of the Mind: A Search for the
Missing Science of Consciousness (1994), and The Large, the Small and the
Human Mind (1997). In addition to his books on consciousness, others he has
written for more general audiences include one with Stephen Hawking entitled
The Nature of Space and Time (1996), The Road to Reality: A Complete Guide
to the Laws of the Universe (2004), and Cycles of Time: An Extraordinary New
View of the Universe (2010).
Gideon A. Rosen, Stuart Professor of Philosophy and chair of the Coun-
cil of the Humanities at Princeton University, specializes in metaphysics,
epistemology, philosophy of mathematics, and moral philosophy. He has
been a visiting professor at the University of Auckland in New Zealand and
held a Mellon Foundation New Directions Fellowship at New York Univer-
sity Law School where he served as the Hauser Fellow in Global Law. Profes-
sor Rosen is the author (with John P. Burgess) of A Subject with No Object:
Strategies for Nominalist Reconstrual in Mathematics (1997), published by
Oxford University Press.
Stewart D. Shapiro is O’Donnell Professor of Philosophy at The Ohio State
University and professorial fellow at the University of St. Andrews. His
research and writing have focused primarily on the philosophy of mathematics,
logic, the philosophy of logic, and the philosophy of language. The recipient
of several fellowships awarded by the National Endowment for the Humanities
and the American Council of Learned Societies, he also has received an
Ohio State Award for Scholarly Achievement and an Ohio State University
Distinguished Scholar Award. Professor Shapiro was an editor of the Journal
of Symbolic Logic and has edited five special issues of journals and three
books, including the Oxford Handbook of the Philosophy of Logic and Math-
ematics (2005). He is also the author of four Oxford University Press books:
Foundations without Foundationalism: A Case for Second-Order Logic (1991
and 2000), Philosophy of Mathematics: Structure and Ontology (1997 and
2000), Thinking about Mathematics: The Philosophy of Mathematics (2000),
and Vagueness in Context (2006). He is writing a new textbook for Oxford
University Press tentatively entitled Logic for Philosophers.
Mark Steiner is a professor of philosophy at The Hebrew University of
Jerusalem. He has specialized in the philosophy of mathematics as part of his
more general attention to the philosophy of science. His work has included
a critical account of Ludwig Wittgenstein’s philosophy of mathematics. He
is the author of Mathematical Knowledge (1975) and The Application of
Mathematics as a Philosophical Problem (1998). His translation from Yiddish
xii L I S T O F C O N T R I B U T O R S
into English of Emune un Apikorses (1948) by Reuven Agushewitz, a
Lithuanian-born Talmudic scholar who attacked the philosophy of materialism
in all its historical versions, was published as Faith and Heresy (2006). He
is now working on a translation of Hume’s Treatise of Human Nature into
Hebrew.
Introduction
John Polkinghorne
Is mathematics a highly sophisticated intellectual game in which the adepts

display their skill by tackling invented problems, or are mathematicians
engaged in acts of discovery as they explore an independent realm of math-
ematical reality? Why does this seemingly abstract discipline provide the key
to unlocking the deep secrets of the physical universe? How one answers
these questions will significantly influence metaphysical thinking about reality.
An interdisciplinary Symposium composed of mathematicians, physicists and
philosophers met twice, at Castel Gandolfo and in Cambridge, to address these
issues. This volume presents the considered form of the contributions that each
participant made to the vigorous discussions that took place. Every effort has
been made to strike a balance between the precision of thought required for
such an enterprise and a reasonable degree of accessibility for a non-specialist
reader prepared to make some effort to engage with the issues.
Peter Lipton, Professor of the Philosophy of Science at Cambridge Uni-
versity, was a valued contributor to our first meeting, and we were all greatly
saddened by his untimely death before we met again. It is the unanimous wish
of all of us involved in the project to dedicate this book to the grateful memory
of a fine scholar and a courteous and stimulating colleague.
The first two chapters are written by mathematicians, Timothy Gowers and
Marcus du Sautoy. They are able to draw on their long and fruitful experience
of doing mathematics. Gowers pays particular attention to how the words
‘invention’ and ‘discovery’ are actually used in the mathematical community.
He concludes that ‘discovery’ seems appropriate when there is essentially a
sole route of argument leading to a significant conclusion, while ‘invention’
is preferred when several distinct lines of argument are available. du Sautoy
describes an incident of insight arising in a flash of inspiration, an experience
which he finds carries the conviction that what had been discerned was ‘already
there’, waiting to be found.
The next two chapters are written by mathematical physicists, John
Polkinghorne and Roger Penrose. Polkinghorne seeks to defend mathematical
2 M E A N I N G I N M AT H E M AT I C S
realism by a variety of arguments, ranging from Gödelian incompleteness to
the evolution of hominid mathematical ability. Both physicists attribute signif-
icance to the role that mathematics plays in affording a strategy for discovery
in their subject. Penrose appeals to Gödelian incompleteness as signifying that
conscious thought is more than neural computation.
The remaining chapters are written by philosophers. Peter Lipton’s chapter
is sadly confined to the short paper that he contributed to the initial meeting of
the Symposium. It discusses the concepts of knowledge, understanding and
explanation, and emphasises the differences he sees between their applica-
tion in science and in mathematics. Stewart Shapiro helpfully provides an
Addendum indicating some ways in which this discussion might be further
amplified. Mary Leng denies that the feeling of discovery to which so many
mathematicians testify necessarily leads to a Platonic view of mathemati-
cal reality. Instead, she suggests it can be understood as arising from the
exploration of logical necessity. Michael Detlefson gives an extensive survey
of approaches, both ancient and modern, to the debate about invention or
discovery. He offers a careful critique of Kurt Gödel’s famous analogy of
mathematical ‘perception’ to sense perception. Stewart Shapiro considers the
argument that mathematics is a human activity, deriving its conventions from
human choices. A key concept for him is ‘cognitive command’, illustrated by
the phenomenon of the necessary agreement between the results of different
persons doing the same calculation. He sees this as an encouragement to taking
the discovery point of view. Gideon Rosen explores the idea that the status of
mathematics corresponds to what he calls ‘qualified realism’. He characterizes
this judgement as a verdict amounting to mathematics being ‘metaphysically
second rate’, because of its dependence on more fundamental logical facts.
Finally, Mark Steiner points us to Descartes rather than Plato. He stresses the
fact that mathematics seems to offer ‘surplus value’, allowing mathematicians
to get out of the axioms more than seems to have been put in. (Mathematicians
themselves call this the quality of being ‘deep’.)
A feature of the Symposium was the liveliness and penetration of its dis-
cussions. The participants wish to convey to the readers of this book something
of the flavour of this experience, and so each has formulated a brief comment
to be attached to the chapter of one of the other participants. We believe that
these responses are an important part of reporting what was a stimulating and
challenging Symposium.
The meetings of the Symposium were supported by the John Templeton
Foundation and all participants wish to express their gratitude for this gen-
erosity. In particular, we were greatly helped by the organizing skill and keen
interest of Dr. Mary Ann Meyers of the Foundation, to whom we offer our
special thanks.
1
Is mathematics discovered or
invented?
Timothy Gowers
The title of this chapter is a famous question. Indeed, perhaps it is a little too
famous: it has been asked over and over again, and it is not clear what would
constitute a satisfactory answer. However, I was asked to address it during the
discussions that led to this volume, and since most of the participants in those
discussions were not research mathematicians, I was in particular asked to give
a mathematician’s perspective on it.
One reason for the appeal of the question seems to be that people can use
it to support their philosophical views. If mathematics is discovered, then it
would appear that there is something out there that mathematicians are discov-
ering, which in turn would appear to lend support to a Platonist conception of
mathematics, whereas if it is invented, then that might seem to be an argument
in favour of a non-realist view of mathematical objects and mathematical truth.
But before a conclusion like that can be drawn, the argument needs to be
fleshed out in detail. First, one must be very clear what it means to say that
some piece of mathematics has been discovered, and then one must explain,
using that meaning, why a Platonist conclusion follows. I do not myself believe
that this programme can be carried out, but one can at least make a start on
it by trying to explain the incontestable fact that almost all mathematicians
who successfully prove theorems feel as though they are making discoveries.
It is possible to think about this question in a non-philosophical way, which
is what I shall try to do. For instance, I shall consider whether there is an
identifiable distinction between parts of mathematics that feel like discoveries
and parts that feel like inventions. This is partly a psychological question and
partly a question about whether there are objective properties of mathematical
statements that explain how they are perceived. The argument in favour of
Platonism only needs some of mathematics to be discovered: if it turns out that
there are two broad kinds of mathematics, then perhaps one can understand
the distinction and formulate more precisely what mathematical discovery (as
opposed to the mere producing of mathematics) is.
As the etymology of the word ‘discover’ suggests, we normally talk of
discovery when we find something that was, unbeknownst to us, already there.
For example, Columbus is said to have discovered America (even if one can
question that statement for other reasons), and Tutankhamun’s tomb was dis-
covered by Howard Carter in 1922. We say this even when we cannot directly
observe what has been discovered: for instance, J. J. Thompson is famous as
the discoverer of the electron. Of greater relevance to mathematics is the dis-
covery of facts: we discover that something is the case. For example, it would
make perfectly good sense to say that Bernstein and Woodward discovered (or
contributed to the discovery) that Nixon was linked to the Watergate burglary.
In all these cases, we have some phenomenon, or fact, that is brought
to our attention by the discovery. So one might ask whether this transition
from unknown to known could serve as a definition of discovery. But a few
examples show that there is a little more to it than that. For instance, an
amusing fact, known to people who like doing cryptic crosswords, is that
the words ‘carthorse’ and ‘orchestra’ are anagrams. I presume that somebody
somewhere was the first person to notice this fact, but I am inclined to call it an
observation (hence my use of the word ‘notice’) rather than a discovery. Why
is this? Perhaps it is because the words ‘carthorse’ and ‘orchestra’ were there
under our noses all the time and what has been spotted is a simple relationship
between them. But why could we not say that the relationship is discovered
even if the words were familiar? Another possible explanation is that once the
relationship is pointed out, one can very easily verify that it holds: you don’t
have to travel to America or Egypt, or do a delicate scientific experiment, or
get access to secret documents.
As far as evidence for Platonism is concerned, the distinction between
discovery and observation is not especially important: if you notice something,
then that something must have been there for you to notice, just as if you
discover it then it must have been there for you to discover. So let us think of
observation as a mild kind of discovery rather than as a fundamentally different
phenomenon.
How about invention? What kinds of things do we invent? Machines are
an obvious example: we talk of the invention of the steam engine, or the
aeroplane, or the mobile phone. We also invent games: for instance, the British
invented cricket—and more to the point, that is an appropriate way of saying
what happened. Art supplies us with a more interesting example. One would
never talk of a single work of art being invented, but it does seem to be
possible to invent a style or a technique. For example, Picasso did not invent
Les Desmoiselles d’Avignon, but he and Braque are credited with inventing
cubism.
A common theme that emerges from these examples is that what we
invent tends not to be individual objects: rather, we invent general methods
for producing objects. When we talk of the invention of the steam engine, we
are not talking about one particular instance of steam-enginehood, but rather of
the idea—that a clever arrangement of steam, pistons, etc. can be used to drive
IS M AT H E M AT I C S D I S C O V E R E D O R I N V E N T E D ? 5
machines—that led to the building of many steam engines. Similarly, cricket

is a set of rules that has led to many games of cricket, and cubism is a general
idea that led to the painting of many cubist pictures.
If somebody wants to argue that the fact of mathematical discovery is
evidence for a Platonist view of mathematics, then what they will be trying to
show is that certain abstract entities have an independent existence, and certain
facts about those entities are true for much the same sort of reason that certain
facts about concrete entities are true. For instance, the statement ‘There are
infinitely many prime numbers’ is true, according to this view, because there
really are infinitely many natural numbers out there, and it really is the case
that infinitely many of them are prime.
A small remark one could make here is that it is also possible to use the
concept of invention as an argument in favour of an independent existence for
abstract concepts. Indeed, our examples of invention all involve abstraction in
a crucial way: ‘the steam engine’, as we have just noted, is an abstract concept,
as are the rules of cricket. Cubism is a more problematic example as it is less
precisely defined, but it is undoubtedly abstract rather than concrete. Why do
we not say that these abstract concepts are brought into existence when we
invent them?
One reason is that we feel that independently existing abstract concepts
should be timeless. So we do not like the idea that when the British invented
the rules of cricket, they reached out into the abstract realm and brought the
rules into existence. A more appealing picture would be that they selected
the rules of cricket from a vast ‘rule space’ that consists of all possible sets
of rules (most of which give rise to terrible games). A drawback with this
second picture is that it fills up the abstract realm with a great deal of junk, but
perhaps it really is like that. For example, it is supposed to contain all the real
numbers, all but countably many of which are undefinable.
Another argument against the idea that one brings an abstract concept
into existence when one invents it is that the concepts that we invent are not
fundamental enough: they tend to be methods for dealing with other objects,
either abstract or concrete, that are much simpler. For example, the rules of
cricket describe constraints on a set of procedures that are carried out by 22
players, a ball and two wickets. From an ontological point of view, the players,
ball and wickets seem more secure than the constraints on how they behave.
Earlier, I commented that we do not normally talk of inventing a single
work of art. However, we do not discover it either: a commonly used word for
what we do would be ‘create’. And most people, if asked, would say that this
kind of creation has more in common with invention than with discovery, just
as observation has more in common with discovery than with invention.
Why is this? Well, in both cases what is brought into existence has many
arbitrary features: if we could turn the clock back to just before cricket was
invented and run the world all over again, it is likely that we would see the
invention of a similar game, but unlikely that its rules would be identical to
those of the actual game of cricket. (One might object that if the laws of physics
are deterministic, then the world would develop precisely as it did the first
time. In that case, one could make a few small random changes before the
rerun.) Similarly, if somebody had accidentally destroyed Les Desmoiselles
d’Avignon just after Picasso started work on it, forcing him to start again, it is
likely that he would have produced a similar but perceptibly different painting.
By contrast, if Columbus had not existed, then somebody else would have
discovered America and not just some huge land mass of a broadly similar kind
on the other side of the Atlantic. And the fact that ‘carthorse’ and ‘orchestra’
are anagrams is independent of who was the first to observe it.
With these thoughts in mind, let us turn to mathematics. Again, it will help
to look at some examples of what people typically say about various famous
parts of the subject. Let me list some discoveries, some observations and some
inventions. (I cannot think of circumstances where I would definitely want to
say that a piece of mathematics was created.) Later I will try to justify why
each item is described in the way it is.
A few well-known discoveries are the formula for the quadratic,
the absence of a similar formula for the quintic, the Monster group, and the
fact that there are infinitely many primes. A few observations are that the
number of primes less than 100 is 25, that the last digits of the powers of
3 form the sequence 3, 9, 7, 1, 3, 9, 7, 1, . . . , and that the number 10001 fac-
torizes as 73 times 137. An intermediate case is the fact that if you define
an infinite sequence z0 , z1 , z2 , . . . of complex numbers by setting z0 = 0 and
zn = z2n−1 + C for every n > 0, then the set of all complex numbers C for
which the sequence does not tend to infinity, now called the Mandelbrot
set, has a remarkably complicated structure. (I regard this as intermediate
because, although Mandelbrot and others stumbled on it almost by accident,
it has turned out to be an object of fundamental importance in the theory of
dynamical systems.)
On the other side, it is often said that Newton and Leibniz independently
invented the calculus. (I planned to include this example, and was heartened
when, quite by coincidence, on the day that I am writing this paragraph,
there was a plug for a radio programme about their priority dispute, and the
word ‘invented’ was indeed used.) One also sometimes talks of mathematical
theories (as opposed to theorems) being invented: it does not sound ridiculous
to say that Grothendieck invented the theory of schemes, though one might
equally well say ‘introduced’ or ‘developed’. Similarly, any of these three
words would be appropriate for describing what Cohen did to the method of
forcing, which he used to prove the independence of the continuum hypoth-
esis. From our point of view, what is interesting is that the words ‘invent’,
‘introduce’ and ‘develop’ all carry with them the suggestion that some general
technique is brought into being.
A mathematical object about which there might be some dispute is the
number i, or more generally the complex number system. Were complex num-
bers discovered or invented? Or rather, would mathematicians normally refer
to the arrival of complex numbers into mathematics using a discovery-type
word or an invention-type word? If you type the phrases ‘complex numbers

were invented’ and ‘complex numbers were discovered’ into Google, you get
approximately the same number of hits (between 4500 and 5000 in both cases),
so there appears to be no clear answer. But this too is a useful piece of data.
A similar example is non-Euclidean geometry, though here ‘discovery of non-
Euclidean geometry’ outnumbers ‘invention of non-Euclidean geometry’ by a
ratio of about 3 to 1.
Another case that is not clear-cut is that of proofs: are they discovered or
invented? Sometimes a proof seems so natural—mathematicians often talk of
‘the right proof’ of a statement, meaning not that it is the only correct proof
but that it is the one proof that truly explains why the statement is true—that
the word ‘discover’ is the obvious word to use. But sometimes it feels more
appropriate to say something like, ‘Conjecture 2.5 was first proved in 1990, but
in 2002 Smith came up with an ingenious and surprisingly short argument that
actually establishes a slightly more general result.’ One could say ‘discovered’
instead of ‘came up with’ in that sentence, but the latter captures better the idea
that Smith’s argument was just one of many that there might have been, and
that Smith did not simply stumble on it by accident.
Let us take stock at this point, and see whether we can explain what it
is about a piece of mathematics that causes us to put it into one of the three
categories: discovered, invented, or not clearly either.
The non-mathematical examples suggested that discoveries and observa-
tions were usually of objects or facts over which the discoverer had no control,
whereas inventions and creations were of objects or procedures with many fea-
tures that could be chosen by the inventor or creator. We also drew some more
refined, but less important, distinctions within each class. A discovery tended
to be more notable than an observation and less easy to verify afterwards. And
inventions tended to be more general than creations.
Do these distinctions continue to hold in much the same form when we
come to talk about mathematics? I claimed earlier that the formula for the
quadratic was discovered, and when I try out the phrase ‘the invention of the
formula for the quadratic’, I find that I do not like it, for exactly
√ the reason
that the solutions of ax2 + bx + c are the numbers (−b ± b2 − 4ac)/2a.
Whoever first derived that formula did not have any choice about what the
formula would eventually be. It is of course possible to notate the formula
differently, but that is another matter. I do not want to get bogged down in
a discussion of what it means for two formulae to be ‘essentially the same’,
so let me simply say that the formula itself was a discovery but that different
people have come up with different ways of expressing it. However, this kind
of concern will reappear when we look at other examples.
The insolubility of the quintic is another straightforward example. It is
insoluble by radicals, and nothing Abel did could have changed that. So his
famous theorem was a discovery. However, aspects of his proof would be
regarded as invention—there have subsequently been very different looking
proofs. This is particularly clear with the closely related work of Galois, who
is credited with the invention of group theory. (The phrase ‘invention of group
theory’ has 40,300 entries in Google, compared with 10 for ‘discovery of group
theory’.)
The Monster group is a more interesting case. It first entered the mathemat-
ical scene when Fischer and Griess predicted its existence in 1973. But what
does that mean? If they could refer to the Monster group at all, then does that
not imply that it existed? The answer is simple: they predicted that a group with
certain remarkable properties (one of which is its huge size—hence the name)
existed and was unique. So to say ‘I believe that the Monster group exists’ was
shorthand for ‘I believe that there exists a group with these amazing properties’
and the name ‘Monster group’ was referring to a hypothetical entity.
The existence and uniqueness of the Monster group were indeed proved,
though not until 1982 and 1990, respectively, and it is not quite clear whether
we should regard this mathematical advance as a discovery or an invention. If
we ignore the story and condense 17 years to an instant, then it is tempting
to say that the Monster group was there all along until it was discovered by
group theorists. Perhaps one could even add a little detail: back in 1973 people
started to have reason to suppose that it existed, and they finally bumped into
it in 1982.
But how did this ‘bumping’ take place? Griess did not prove in some
indirect way that the Monster group had to exist (though such proofs are
possible in mathematics). Rather, he constructed the group. Here, I am using
the word that all mathematicians would use. To construct it, he constructed an
auxiliary object, a complicated algebraic structure now known as the Griess
algebra, and showed that the symmetries of this algebra formed a group with
the desired properties. However, this is not the only way of obtaining the
Monster group: there are other constructions that give rise to groups that have
the same properties, and hence, by the uniqueness result, are isomorphic to it.
So it seems that Griess had some control over the process by which he built the
Monster group, even if what he ended up building was determined in advance.
Interestingly, the phrase ‘construction of the Monster group’ is much more
popular on Google than the phrase ‘discovery of the Monster group’ (8290
to 9), but if you change it to ‘the construction of the Monster group’ then it
becomes much less popular (6 entries), reflecting the fact that there are many
different constructions.
Another question one might ask is this. If we do decide to talk about the
discovery of the Monster group, are we talking about the discovery of an
object, the Monster group, or of a fact, the fact that there exists a group with
certain properties and that that group is unique? Certainly, the second is a better
description of the work that the group theorists involved actually did, and the
word ‘construct’ is a better word than ‘discover’ at describing how they proved
the existence part of this statement.
The other discoveries and observations listed earlier appear to be more
straightforward, so let us turn to the examples on the invention side.
A straightforward use of the word ‘invention’ in mathematics is to refer to

the way general theories and techniques come into being. This certainly covers
the example of the calculus, which is not an object, or a single fact, but rather
a large collection of facts and methods that greatly increase your mathematical
power when you are familiar with them. It also covers Cohen’s technique of
forcing: again, there are theorems involved, but what is truly interesting about
forcing is that it is a general and adaptable method for proving independence
statements in set theory.
I suggested earlier that inventors should have some control over what they
invent. That applies to these examples: there is no clear criterion that says
which mathematical statements are part of the calculus, and there are many
ways of presenting the theory of forcing (and, as I mentioned earlier, many
generalizations, modifications and extensions of Cohen’s original ideas).
How about the complex number system? At first sight this does not look at
all like an invention. After all, it is provably unique (up to the isomorphism that
sends a + bi to a − bi), and it is an object rather than a theory or a technique.
So why do people sometimes call it an invention, or at the very least feel a little
uneasy about calling it a discovery?
I do not have a complete answer to this question, but I suspect that the
reason it is a somewhat difficult example is similar to the reason that the
Monster group is difficult, which is that one can ‘construct’ the complex
numbers in more than one way. One approach is to use something like the way
they were constructed historically (my knowledge of the history is very patchy,
so I shall not say how close the resemblance is). One simply introduces a new
symbol, i, and declares that it behaves much like a real number, obeying all the
usual algebraic rules, and has the additional property that i2 = −1. From this
one can deduce that
(a + bi)(c + di) = ac + bci + adi + bdi2 = (ad − bd) + (ad + bc)i
and many other facts that can be used to build up the theory of complex
numbers. A second approach, which was introduced much later in order to
demonstrate that the complex number system was consistent if the real number
system was, is to define a complex number to be an ordered pair (a, b) of real
numbers, and to stipulate that addition and multiplication of these ordered pairs
are given by the following rules:
(a, b) + (c, d) = (a + c, b + d)
(a, b)(c, d) = (ac − bd, ad + bc)
This second method is often used in university courses that build up the number
systems rigorously. One proves that these ordered pairs form a field under the
two given operations, and finally one says, ‘From now on I shall write a + bi
instead of (a, b).’
Another reason for our ambivalence about the complex numbers is that
they feel less real than real numbers. (Of course, the names given to these
numbers reflect this rather unsubtly.) We can directly relate the real numbers
to quantities such as time, mass, length, temperature, and so on (though for this
we never need the infinite precision of the real number system), so it feels as
though they have an independent existence that we observe. But we do not run
into the complex numbers in that way. Rather, we play what feels like a sort of
game—imagine what would happen if −1 did have a square root.
But why in that case do we not feel happy just to say that the complex
numbers were invented? The reason is that the game is much more interesting
than we had any right to expect, and it has had a huge influence even on
those parts of mathematics that are about real numbers or even integers. It
is as though after our one small act of inventing i, the game took over and
we lost control of the consequences. (Another example of this phenomenon is
Conway’s famous game of Life. He devised a few simple rules, by a process
that one would surely want to regard as closer to invention than discovery, but
once he had done so he found that he had created a world full of unexpected
phenomena that he had not put there, so to speak. Indeed, most of them were
discovered—to use the obvious word—by other people.)
Why is ‘discovery of non-Euclidean geometry’ more popular than ‘inven-
tion of non-Euclidean geometry’? This is an interesting case, because there
are two approaches to the subject, one axiomatic and one concrete. One could
talk about non-Euclidean geometry as the discovery of the remarkable fact
that a different set of axioms, where the parallel postulate is replaced by a
statement that allows a line to have several parallels through any given point,
is consistent. Alternatively, one could think of it as the construction of models
in which those axioms are true. Strictly speaking, one needs the second for the
first, but if one explores in detail the consequences of the axioms and proves
all sorts of interesting theorems without ever reaching a contradiction, that can
be quite impressive evidence for their consistency. It is probably because the
consistency interests us more than the particular choice of model, combined
with the fact that any two models of the hyperbolic plane are isometric, that we
usually call it a discovery. However, Euclidean geometry (wrongly) feels more
‘real’ than hyperbolic geometry, and there is no single model of hyperbolic
geometry that stands out as the most natural one; these two facts may explain
why the word ‘invention’ is sometimes used.
My final example was that of proofs, which I claimed could be discovered
or invented, depending on the nature of the proof. Of course, these are by
no means the only two words or phrases that one might use: some others
are ‘thought of’, ‘found’, ‘came up with’. And often one regards the proof
less as an object than as a process, and focuses on what is proved, as is
shown in sentences such as, ‘After a long struggle, they eventually managed
to prove/establish/show/demonstrate that . . . ’ Proofs illustrate once again the
general point that we use discovery words when the author has less control and
invention words when there are many choices to be made. Where, one might
ask, does the choice come from? This is a fascinating question in itself, but let
me point out just one source of choice and arbitrariness: often a proof requires
one to show that a certain mathematical object or structure exists (either as
the main statement or as some intermediate lemma), and often the object or
structure in question is far from unique.
Before drawing any conclusions from these examples, I would like to
discuss briefly another aspect of the question. I have been looking at it mainly
from a linguistic point of view, but, as I mentioned right at the beginning, it
also has a strong psychological component: when one is doing mathematical
research, it sometimes feels more like discovery and sometimes more like
invention. What is the difference between the two experiences?
Since I am more familiar with myself than with anybody else, let me draw
on my own experience. In the mid 1990s I started on a research project that
has occupied me in one way or another ever since. I was thinking about a
theorem that I felt ought to have a simpler proof than the two that were then
known. Eventually I found one (here I am using the word that comes naturally);
unfortunately it was not simpler, but it gave important new information. The
process of finding this proof felt much more like discovery than invention,
because by the time I reached the end, the structure of the argument included
many elements that I had not even begun to envisage when I started working on
it. Moreover, it became clear that there was a large body of closely related facts
that added up to a coherent and yet-to-be-discovered theory. (At this stage,
they were not proved facts, and not always even precisely stated facts. It was
just clear that ‘something was going on’ that needed to be investigated.) I and
several others have been working to develop this theory, and theorems have
been proved that would not even have been stated as conjectures fifteen years
ago.
Why did this feel like discovery rather than invention? Once again it is
connected with control: I was not selecting the facts I happened to like from
a vast range of possibilities. Rather, certain statements stood out as obviously
natural and important. Now that the theory is more developed, it is less clear
which facts are central and which more peripheral, and for that reason the
enterprise feels as though it has an invention component as well.
A few years earlier, I had a different experience: I found a counterexample
to an old conjecture in the theory of Banach spaces. To do this, I constructed
a complicated Banach space. This felt partly like an invention—I did have
arbitrary choices, and many other counterexamples have subsequently been
found—and partly like a discovery—much of what I did was in response to the
requirements of the problem, and felt like the natural thing to do, and a very
similar example was discovered independently by someone else (and even the
later examples use similar techniques). So this is another complicated situation
to analyse, but the reason it is complicated is simply that the question of how
much control I had is a complicated one.
What conclusion should we draw from all these examples and from how
we naturally seem to regard them? First, it is clear that the question with which
we began is rather artificial. For a start, the idea that either all of mathematics
is discovered or all of mathematics is invented is ridiculous. But even if we
look at the origins of individual pieces of mathematics, we are not forced to
use the word ‘discover’ or ‘invent’, and we very often don’t.
Nevertheless, there does seem to be a spectrum of possibilities, with some
parts of mathematics feeling more like discoveries and others more like inven-
tions. It is not always easy to say which are which, but there does seem to be
one feature that correlates strongly with whether we prefer to use a discovery-
type word or an invention-type word. That feature is the control that we have
over what is produced. This, as I have argued, even helps to explain why the
doubtful cases are doubtful.
If this is correct (perhaps after some refinement), what philosophical con-
sequences can we draw from it? I suggested at the beginning that the answer
to the question did not have any bearing on questions such as ‘Do numbers
exist?’ or ‘Are mathematical statements true because the objects they mention
really do relate to each other in the ways described?’ My reason for that
suggestion is that pieces of mathematics have objective features that explain
how much control we have over them. For instance, as I mentioned earlier, the
proof of an existential statement may well be far from unique, for the simple
reason that there may be many objects with the required properties. But this
is a straightforward mathematical phenomenon. One could accept my analysis
and believe that the objects in question ‘really exist’, or one could view the
statements that they exist as moves in games played with marks on paper, or
one could regard the objects as convenient fictions. The fact that some parts of
mathematics are unexpected and others not, that some solutions are unique and
others multiple, that some proofs are obvious and others take a huge amount
of work to produce—all these have a bearing on how we describe the process
of mathematical production and all of them are entirely independent of one’s
philosophical position.
Comment on Timothy Gowers’ ‘Is
mathematics discovered or invented?’
Gideon Rosen
In a nearby possible world Timothy Gowers is not the distinguished

mathematician that he is in our world, but rather a 1950s-style ordinary lan-
guage philosopher. In his contribution to this volume he approaches his title
question—‘Is mathematics discovered or invented?’—by attending rather care-
fully to the ways in which mathematicians (and the variously informed hordes
whose musings are lodged in Google’s database) actually use these words in
application to various parts of mathematics. Gowers’ conclusion is (roughly)
that the rhetoric of ‘discovery’ strikes us as apt when the mathematician has
no significant choice about how he does what he does, whereas we are inclined
to speak of ‘invention’ or perhaps ‘construction’ when there are many ways to
perform the task at hand and the mathematician has some control over how he
does it.
Gowers is keen to insist that the distinctions that interest him are indepen-
dent of one’s views about the metaphysics of mathematics.
One could accept my analysis and believe that the objects in question “really
exist”, or one could view the statements that they exist as moves in games
played with marks on paper, or one could regard the objects as convenient
fictions. The fact that some parts of mathematics are unexpected and others
not, that some solutions are unique and others multiple, that some proofs are
obvious and others take a huge amount of work to produce—all these have a
bearing on how we describe the process of mathematical production and all
of them are entirely independent of one’s philosophical position. (p. 12)
This strikes me as exactly right, but it raises a question that Gowers does not
address. Gowers has described the conditions under which mathematicians
are inclined to say that some achievement amounts to a discovery or an
invention, and also the conditions under which an achievement is likely to
feel like a discovery or an invention to those whose achievement it is. But how
seriously should we take these linguistic and psychological observations? As
philosophers have often noted, it is one thing to chart the circumstances under
which we are inclined to say this or that, another to identify the conditions
under which it is literally correct to say this or that. So let us grant that
mathematicians agree in their classification of some episode as a matter of
(say) invention. Does that entail or even suggest that this episode was in fact a
matter of invention? Or is this rather a mere manner of speaking that it would
be a mistake to take too seriously?
I believe that this question has different answers in different cases. As
Gowers notes, we speak of the invention/discovery of many kinds of thing:
theories, theorems, proofs and proof techniques, but also mathematical objects
of various sorts (numbers, number systems). We can say that Cantor invented
the theory of transfinite numbers, but we are much less likely to say that Cantor
invented the transfinite numbers themselves. Let’s focus first on the rhetoric
of invention/construction as applied to mathematical objects. Here Gowers
discusses the case of the Monster group, an enormous finite group whose
existence and uniqueness were proved in 1982 and 1990, respectively. The
linguistic evidence suggest that mathematicians are more inclined to speak of
the ‘construction’ of the Monster group than of its ‘discovery’, and Gowers’
account explains this. The proof of the existence of the Monster group is not
unique: many examples may be adduced to establish the existential theorem
that a group with the relevant properties exists, even though (as it happens)
every such example is isomorphic to every other. But is there any reason to
take the imagery of construction seriously in this case? In my view it is a
non-negotiable feature of the literal use of this idiom that if a thing has been
invented or constructed, it did not exist before it was invented and would
not have existed if it had not been invented. By contrast, when a thing is
discovered, it must exist prior to (or at least independently of) the episode
of discovery. But as I think Gowers would agree, it would be quite odd to say
that before 1982, the Monster group did not exist. If this were the right thing
to say, then when Griess first asked himself the question, ‘Does the Monster
exist?’ the answer should have been obvious: ‘Not yet, but maybe someday.’
But in fact no one speaks of mathematical objects in this way. I am therefore
tempted to conclude that even if Gowers is right about the conditions under
which we are inclined to reach for the language of invention or construction
in connection with mathematical objects, it would be a mistake to take this
language literally in this connection.
Things are otherwise when it comes to mathematical theories—especially
large theories like the calculus. If someone had asked in (say) 1650 whether
there existed a powerful battery of algebraic techniques for calculating the area
bounded by a curve and the line tangent to a curve at a point, and a deep theory
justifying these techniques and displaying their connections, the answer might
well have been, ‘Not yet, but maybe someday.’ Moreover, it seems equally
natural to say that if no one had ever managed to write down such a theory, then
the calculus, as we know it, would not exist. Theories of this sort thus appear to
belong to the same ontological category as novels and poems and philosophical
IS M AT H E M AT I C S D I S C O V E R E D O R I N V E N T E D ? C O M M E N T 15
treatises. Such things are abstract artifacts: abstract entities that come into
existence when someone produces a concrete representation of them for the
first time. In these cases, I see no reason not to take the rhetoric of invention
seriously as a sober and literal account of the underlying metaphysics.
Gowers makes no claims of this sort, but I wonder, however, whether he
would agree with my assumption that unless we are prepared to say that the
invented item did not exist prior to its invention, we should regard claims of
invention (construction, creation, etc.) in mathematics as metaphorical. We
might then take Gowers’ careful account of the conditions under which we are
inclined to deploy the metaphor as an account of the sober and metaphysically
neutral truth that underlies it.
2
Exploring the mathematical library
of Babel
Marcus du Sautoy
I’m a mathematician, not a philosopher. My job is to prove new theorems.

To discover new truths about the numbers we count with. To create new
symmetrical objects. To find new connections between disparate parts of the
mathematical landscape.
Yet contained in my job description are a whole bunch of words that raise
important questions about what mathematics is and how it relates to the physi-
cal and mental world we inhabit. ‘Create’, ‘discover’, ‘proof’, ‘truth’. All very
emotive words. And every mathematician at some point will find themselves
contemplating whether a new mathematical breakthrough they’ve just made
is an act of creation or an act of discovery. Is mathematics an objective or
subjective activity? Do mathematical objects exist?
The only way for me to engage with these questions is to analyse what I
think I do when I do mathematics. So I’ve chosen an episode from my working
life to help me explore some of these issues. (More details of this discovery
can be found in du Sautoy, 2009.)
One of my proudest moments as a mathematician was constructing a new
symmetrical object whose subgroup structure is related to counting the number
of solutions modulo p of an elliptic curve. Finding solutions to elliptic curves is
one of the toughest problems on the mathematical books. An elliptic curve is an
equation like y2 = x3 − x (or more generally a quadratic in y equal to a cubic
in x). One of the million-dollar-prize problems offered by the Clay Institute,
called the Birch and Swinnerton–Dyer Conjecture, seeks to understand when
one of these equations has infinitely many solutions where both x and y are
fractions.
I constructed this symmetrical object whose structure encoded this impor-
tant question of solving equations while working at the Max Planck Institute
in Bonn. A mathematical theorem that I’d proved with a colleague in Germany
hinted that such symmetrical objects might exist, but until such a symmetry
Fig. 2.1 The construction in my notebook of a new symmetrical object.
group was constructed that demonstrated the connection it might just be an

illusion. That evening sitting in Bonn was one of those moments that mathe-
maticians often talk about, when a flash of inspiration hits you. I wrote down
the construction of how these symmetries of the new object should interact
with each other on the yellow legal pad that is the palette for my mathematical
musings.
It felt right. It took a few more days to really prove what I thought it did.
But once the details were fleshed out, this new object revealed a connection
between the world of symmetry and the world of arithmetic geometry that had
not been seen before.
Of course, when I say I constructed this symmetrical object, I didn’t physi-
cally build it. This is an object of the mind living in the abstract world of math-
ematics. It isn’t like the person who first carved out an icosahedron with its 20
triangular faces or the Moorish artist in the Alhambra, Granada, who found a
new way to cover the walls with symmetrical tiles. A physical representation of
the object I discovered would only exist in some high-dimensional space. And
even so, these representations are only expressions of the underlying group of
symmetries. Both the rotations of the icosahedron and the dodecahedron are
examples of two objects with the same underlying symmetry group called A5 .
Similarly, these two designs found in the Alhambra, although physically very
different, have identical underlying groups of symmetries.
Just as the number ‘three’ captures the identity of a collection which
has three objects, whether it be three apples or three kangaroos, the nam-
ing of the symmetry 632 has abstracted the symmetrical identity shared by
these two walls in the Alhambra. The abstract symmetry group is described
by giving names to each of the symmetries and then explaining how these
symmetries interact with each other when you do one symmetry followed by
another.
EXPLORING T H E M AT H E M AT I C A L L I B R A R Y O F B A B E L 19
Fig. 2.2 Two walls in the Alhambra with the same group of symmetries called 632.
What I’d ‘constructed’ that evening in Bonn was an abstract symmetrical

object whose symmetries interact in such a way to produce interesting new
connections with elliptic curves. It certainly doesn’t exist in the physical world,
yet when you spend enough time in the mathematical world it has a reality that
is akin to handling a dodecahedron or tiling a wall in the Alhambra.
I’ve been careful to avoid using the word ‘create’ while describing the
episode above but I had to fight myself not to write the word. Because con-
structing this new group of symmetries certainly felt like an act of creation.
I experienced the strong sense that the scribblings I penned on my yellow pad
brought into existence something new, something that didn’t exist before I’d
described its contours. It was through my act of imagination that this thing
emerged. It required my agency to realize the existence of this object. It wasn’t
something that would naturally evolve without me being present. I provided it
with the push that gave it life.
The creative side of mathematics is one that many mathematicians talk
about. It is one of the reasons that drew me to mathematics rather than the
other sciences, which I felt were more about observation. When I was at
school, I was very interested in music; I was learning the trumpet, I enjoyed
the theatre, enjoyed reading. Science hadn’t really captured my imagination.
But then around the age of 13 my mathematics teacher took me aside after
one lesson: ‘I think you should find out what mathematics is really about.
Mathematics is not about all the multiplication tables and long division we
do in the classroom. It is actually much more exciting than that and I think
you might enjoy seeing the bigger picture.’ He gave me the names of some
books that he thought I might enjoy and would open up what this world of
mathematics was all about.
One of the books was A Mathematician’s Apology by G. H. Hardy (1940).
It had a big impact on me. As I read Hardy’s book, there were sentences which
revealed to me that mathematics shared a lot in common with the creative
arts. It seemed to be compatible with things I loved doing: languages, music,
reading. Here for example is Hardy, writing about being a mathematician:
‘A mathematician, like a painter or a poet, is a maker of patterns. If his patterns
are more permanent than theirs, it is because they are made with ideas.’ Later
he writes: ‘The mathematician’s patterns, like the painter’s or the poet’s, must
be beautiful; the ideas, like the colours or the words, must fit together in an
harmonious way. Beauty is the first test: there is no permanent place in the
world for ugly mathematics.’ For Hardy, mathematics was a creative art, not
a useful science. ‘The “real” mathematics of the “real” mathematicians, the
mathematics of Fermat and Euler and Gauss and Abel and Riemann, is almost
wholly “useless” (and this is true of “applied” as of “pure” mathematics). It is
not possible to justify the life of any genuine professional mathematician on
the ground of the “utility” of his work.’
The construction of my group of symmetries certainly wasn’t motivated
by utility. It was the creation of something that appealed to my sense of
aesthetics. It was surprising. It had an unexpected twist. Like a theme in
a piece of music, it mutated during the course of the proof into something
quite different. I suppose part of my motivation for constructing this group
of symmetries was a certain mathematical utility. It might ultimately help us
understand elliptic curves better; it gives a new perspective on the complexity
of classifying p-groups. But it still felt like a creative act not forced on me by
external factors outside my control.
And yet . . . wasn’t this mathematical object just sitting out there waiting for
someone to notice it? Wasn’t my moment in Bonn just an act of discovery? If
it wasn’t me who discovered it, wouldn’t someone else have eventually come
to the same realization? I was scrambling round the mathematical landscape
and uncovered this symmetrical object. Wasn’t it there all along, waiting to be
carved out of the ground? Why was it any different to the first scientist who
discovered gold or the first astronomer to spot the planet Neptune.
Here’s Hardy in a completely different frame of mind from his talk of
the creativity of mathematics: ‘I believe that mathematical reality lies out-
side us, that our function is to discover or observe it and that the theorems
which we prove and which we describe grandiloquently as our “creations”
are simply our notes of our observations.’ It sums up the schizophrenic rela-
tionship that I think all mathematicians have towards their work. No creative
accountancy can make a prime number divisible. As Hardy declared: ‘317 is
a prime not because we think so, or because our minds are shaped in one
way or another, but because it is so, because mathematical reality is built
that way.’
Maybe there is a difference between discovering a new group of symme-
tries and discovering a new element or planet because gold and Neptune have
naturally evolved and didn’t require my agency. But I still feel that if I hadn’t
discovered this symmetry group, then it was lying out there for someone else
to construct. How much is it a product of my imagination? History records
a catalogue of events where mathematical objects were discovered simulta-
neously and independently by different mathematicians. The most famous is
the discovery of non-Euclidean geometries which was made independently
by Gauss, Bolyai and Lobachevsky. Although their notation, descriptions and
explanations might have been quite personal, the object they discovered, a
geometry with triangles whose angles add up to less than 180 degrees, was
the same.
In contrast, one can’t imagine three composers simultaneously composing
‘The Death and the Maiden’ string quartet. That was a creation of Schubert’s
genius, made at the same time as non-Euclidean geometry was first emerging.
But although the piece of music itself is unique and could never be replicated
by another composer, it is striking that moods and changes in genre in music
and the other arts are happening independently and simultaneously. Composers
are discovering new ways of composing, new structures, new possibilities,
often at the same time. Schubert’s quartet marks the beginning of the Romantic
period of musical composition. But he wasn’t the only one exploring the
ideas of restless key modulations and the heightened contrasts of Sturm und
Drang. Contemporary composers I’ve worked with talk of being beaten to the
discovery of an idea, as if composers are equally discovering new structures,
new forms within which to frame their composition.
Perhaps I can make a proposal that explains the feeling of creativity that
I have when I do mathematics. There were many different groups of symme-
tries that I could have written down on my yellow pad that evening in Bonn.
Infinitely many in fact. All I have to do is write down names for the symmetries
and define how they interact and . . . voilà . . . I’ve created/discovered a new
group. There will be a question of whether the group of symmetries has been
constructed before, but I am more interested in focusing on why I was so
excited about the particular group of symmetries I constructed that evening.
I think it is helpful to consider an analogy with a composer or a writer.
I can randomly write down notes on a stave, give the notes different lengths,
different dynamics, and I will have composed a piece of music. Or I can sit
at a typewriter and just bang out strings of letters or words and write a book.
Borges’s The Library of Babel contains every book composed of 25 letters
where each book consists of 410 pages; each page is made up of 40 lines each
consisting of 80 positions. There are of course a lot of books in the library,
251312000 to be precise.
They are all sitting there waiting for an author to discover one. The pos-
sibility of Great Expectations existed out there before Charles Dickens pulled
the book off the shelf. The act of creativity is in singling out this book among
all the possible books to write. And I think that it is the same characteristic that
is involved in doing mathematics and which is often overlooked.
I could write down endless new and original theorems. I could construct
infinitely many new symmetry groups. I could get a computer to churn them
out for me by just applying the rules of logical deduction from each previous
statement. All of them would have an objective truth about them. All of them
are mathematically true statements. But the point is that just as most of the
books in the library of Babel are not interesting, so too most of these new
theorems are banal or without interest.
There is more to mathematics that just generating mathematical truths. The
art in being a mathematician is to single out those logical pathways that have
something special about them. And here I think a sense of aesthetics plays a
key part in making those choices. The reason I wanted to tell everyone about
my discovery of this new group of symmetries is that it was surprising. It was
like a moment in a novel when you think the main character is one thing but
then suddenly he mutates into something quite different.
Maybe the difference between mathematics and the other sciences is that
it is the natural world that is acting as the agent, picking out those things
which have a special quality about them and we, as scientists, then try to
understand why they are so special and selected. Often the answer is ultimately
a mathematical one.
I like the suggestion made elsewhere in this volume that quite often mathe-
matics is valued when you seem to get more out than you put in. The definition
of a group of symmetries looks quite simple. It’s hard to believe that it has led
to the discovery of strange objects like the Monster and E8 .
Cultural and historical context have an effect on the reception and excite-
ment over different mathematical discoveries. Every 21st century mathemati-
cian cares whether there is a zero of the Riemann zeta function sitting off the
critical line. Even though it would be an impressive feat of mathematics, I just
don’t think 21st century mathematicians care if there are odd perfect numbers
or not. That’s why no one is really working hard to prove this fact. In contrast,
the ancient Greeks might have got very excited by the discovery. Of course the
proof might yield exciting new insights about numbers that mathematicians
might value. Does anyone really care whether Fermat’s equations xn + yn = zn
have integer solutions or not? Certainly there aren’t many theorems that started
‘suppose Fermat’s Last Theorem is true then . . . ’ Why the mathematical com-
munity continued to pursue a proof of this theorem is because it was a catalyst
for the discovery of some amazing ideas.
One might try to make the distinction between mathematical and artistic
creations by declaring that mathematics is discovering eternal truths about the
universe. I can’t make a theorem true just because I think it will be beautiful.
If the Riemann Hypothesis turns out to be false it will shatter our sense of how
beautifully we believe the primes are laid out. But there will be nothing we
can do about it. The Riemann Hypothesis is either true or false and there is no
act of creative thinking which is going to alter that. In contrast, one can’t talk
about the objective truth of ‘The Death and the Maiden’ or Great Expectations.
For a start, the works elicit multiple reactions from audiences. Ambiguity is an
important part of creating art. Ambiguity for the mathematician is anathema.
But the creative act involved in doing mathematics is the act of focusing on
asking the question whether the Riemann Hypothesis is true. There are lots
of questions that we can ask about the primes. Why this one is the Holy
Grail is again because it says something very special about the primes. The
connection between the primes and the zeros of the Riemann zeta function
can’t help but bowl you over when you read about it for the first time. It is such
an extraordinary transformation.
Another key point about mathematical discoveries is how integrated they
are across the subject. This integration is often important in how a piece of
mathematics is valued. A mathematical discovery which seems isolated from
the mathematical mainstream, however surprising or beautiful, will probably
not receive the same sort of attention as mathematics that has connections
with other bits of the subject. The fact that the Riemann Hypothesis is so
interconnected with so many other bits of mathematics is one reason this
mathematics is valued. It’s like the Internet. More links and the higher your
mathematical Google ranking.
Perhaps musical and literary creations can survive better in isolation,
although quite often one can only truly appreciate these creations in relation to
what has gone before.
The quest to prove the Riemann Hypothesis raises the interesting question
of whether there is a difference between proving a conjecture and constructing
new mathematical objects. Of course the creativity involved in constructing
a proof that establishes whether the Riemann Hypothesis is true or not is
matched by the creative act of constructing new non-Euclidean geometries. But
there does feel like a difference in the process. It is a bit like being an explorer.
Riemann pointed out a far distant mountain. Those trying to prove the Riemann
Hypothesis are trying to find some route through the mathematical landscape
to arrive at that mountain. Bolyai’s discovery of non-Euclidean geometry is
like the explorer coming across a new island in the middle of the ocean that
has never been seen before.
What about the question of whether mathematical objects really exist?
I certainly am a Platonist at heart. There are some things out there that are
independent of our existence or act of imagining them. Prime numbers, simple
groups, elliptic curves. It is not mathematicians who made these things. But
then maybe I am getting back to the feeling that my group of symmetries is
simply the articulation of a mathematical entity that was there all along. I think
I have some sympathy with Kronecker’s statement: ‘God made the integers; all
the rest is the work of man.’ That’s not to say that the Riemann Hypothesis is
made true or false by the work of man or woman. It is either true or false
that the primes are distributed as Riemann’s Hypothesis predicts. But it’s the
decision to obsess about this question of mathematics and not some other
hypothesis that is the work of man or woman. Again I think that the act of the
mathematician is to tell particular stories about the integers and to single out
those which are genuinely interesting and surprising to other mathematicians.
I would argue that the aesthetic judgement that singles out great mathematics
shares a lot in common with the traits that one is looking for in a great piece of
music. It is very rarely the usefulness of a piece of mathematics that motivates
a mathematician. It is often centuries later that a mathematical discovery ends
up being applied to the real world. Rather, the mathematician is drawn to
mathematics that is full of beauty, elegance and surprise. In a mathematical
proof, themes are established then mutate, interweave, producing surprising
moments of connection. These are traits that for me make exciting music as
well as exciting mathematical arguments.
For a mathematician, the journey of that proof, either carving the pathway
out for the first time or following in someone else’s footsteps, is the essence
of the mathematics, not the stark statement of the theorem being proved.
For example, Fermat discovered the amazing fact that every prime number
that has remainder 1 on division by 4 can always be written as two square
numbers added together. For example 41 is a prime number which certainly
has remainder 1 on division by 4. Fermat’s theorem guarantees that this prime
can be written as the sum of two squares, in this case: 25 + 16 or 5 squared
plus 4 squared.
For a mathematician this theorem is exciting because it connects two very
different sorts of numbers: primes and squares. But for the mathematician it
is reading a proof of why there is this connection which provides the real
pleasure. The kick comes from the moment when you suddenly see why there
is a common thread connecting the squares and these primes. They are like two
different variations on a common theme.
Just in the same way that people have started to quantify what makes good
music by trying to plot its different characteristics, it might be possible to give
some measure of why we regard some proven statements of mathematics as
being worthy of prizes and publication in the Annals of Mathematics while
others are just ignored as uninteresting. Is it to do with a certain complexity
of the proof? Sometimes. Although simplicity is often a guiding light for
a mathematician. The proof of the four-colour-map problem is complex but
not beautiful and doesn’t provide quite that magic ‘ah-ha’ moment when you
suddenly get why four not five colours suffice. The proof of Fermat’s Last
Theorem is extraordinary and complex (and certainly was not what Fermat
couldn’t fit in the margin) but still a mathematician reading it is swept along
by the twists and turns of ideas like a grand Wagnerian opera arriving at the
final Q. E. D., realizing what an inevitable journey Wiles took you on. Another
measure of the significance of the mathematics is related again to this idea of
getting more out than you put in. A third measure is how integrated the result
is with other mathematics that is valued—that mathematical Google rating.
But maybe trying to quantify what makes good mathematics is just as
doomed to failure as trying to measure why Mozart’s music is so magical.
As Hardy writes in A Mathematician’s Apology: ‘It may be hard to define
mathematical beauty, but that is just as true of beauty of any kind—we may
not know quite what we mean by a beautiful poem, but that does not prevent
us from recognizing one when we read it.’
I often feel that the create/discover question shares something with the
nature/nurture debate. How much is a child just a product of their genes? Does
the environment have much more influence on the outcome and characteristics
of a child? The theorems that the mathematician discovers are their children,
their legacy. The birth of a theorem is often preceded by long hard labours.
Their existence is a way of continuing our legacy. The permanence of their
proof is our chance of a bit of immortality. But are these theorems simply a
consequence of the logical framework we work within, like some genetic code
forcing their behaviour and existence? Or is our nurturing of those theorems we
create a function of the culture, the surrounding environment of the mathemat-
ics that exists around us? It’s not a very satisfying answer for a mathematician
who likes things black or white, true or false, proved or disproved, but it’s
probably a little bit of both. But maybe that’s why at the end of all these
philosophical musings mathematicians so often just stick their heads back in
the mathematical sands and continue trekking through its beautiful landscape,
proving new theorems, constructing new mathematical structures, revelling in
its unchanging certainty. This is the job of a mathematician.
Comment on Marcus du Sautoy’s
‘Exploring the mathematical library
of Babel’
Mark Steiner
Professor du Sautoy reconciles the realist and the constructivist positions in the
philosophy of mathematics with a simple, but effective distinction. Structures
describable in mathematical language exist independently of our knowledge;
this is the realist part. The mathematician chooses, from among these struc-
tures, those which are to be called mathematical structures. To be describable
in mathematical language is not yet to be a mathematical structure. Professor
du Sautoy adds that aesthetical considerations play a dominant role in deciding
what is worth investigating, i.e., what is to be called mathematics. This is
exactly the position I took in my book, The Applicability of Mathematics
as a Philosophical Problem (1998). The way I put it is that mathematics is
anthropocentric to the extent that anthropocentric criteria (like aesthetical)
govern what is called mathematics.
What now of Hardy’s view that beautiful mathematics is never ‘useful’,
which du Sautoy quotes approvingly? I do not see any reason for Professor
du Sautoy to accept what is a patently false view, based mainly on wishful
thinking. (Hardy didn’t want mathematics to be used for warfare.) Hardy is so
spectacularly wrong that, on the contrary, many scientists are convinced that
the more beautiful mathematics is, the more applications it has. Hardy wrote,
‘No one has yet discovered any warlike purpose to be served by the theory
of numbers or relativity, and it seems very unlikely that anyone will do so for
many years.’ While the view of some that Albert Einstein invented the atomic
bomb is ludicrous, to say on the other hand that there is no warlike purpose for
the equivalence of mass with energy is equally ludicrous. And as for number
theory, much of the work in the field, I am told, is simply classified, because it
could be used, and is used, in cryptography. If somebody came up with a good
algorithm for factoring large numbers he would probably be arrested.
I leave Professor du Sautoy, with the following challenge: can you think of
an explanation why beautiful mathematics tends to be useful in applications?
3
Mathematical reality
John Polkinghorne
Are mathematicians engaged in acts of discovery or are they merely construct-

ing ingenious intellectual puzzles whose solutions simply afford occupation
and amusement for those whose tastes lie in that direction? Is mathematics just
the painstaking unravelling of a monstrous logical tautology? Or is mathemat-
ics something much more interesting and significant than either of these rather
banal judgements would suggest?
Seeking answers to these questions is not just a way of assessing the dignity
and importance of mathematics itself, for the result of the enquiry promises
also to provide a significant source of insight into the discussion of wider
and deeper philosophical issues. The status of mathematics bears upon an
answer to the fundamental metaphysical question, ‘What are the dimensions
of reality?’ Do they extend beyond the frontiers of a domain that is capable of
being fully described simply in terms of exchanges of energy between material
constituents, located within the arena of space-time? For the materialist, the
latter is indeed the true extent of reality, and all other human talk, such as
that employing mental or axiological categories, amounts to no more than
convenient manners of speaking about epiphenomena of the material. Or, on
the contrary, is it the case that true ontological adequacy requires that much
more be said than physicalism can articulate?
The issue of the nature of mathematical entities provides a convenient test
case for probing this general question. Particularly helpful as an introduction
to the considerations involved in pursuing the matter is the published report
of an extended conversation between two distinguished French savants, Jean-
Pierre Changeux, a molecular neurobiologist and a resolute materialist, and
Alain Connes, a mathematician and a firm believer in mathematical reality
(Changeux and Connes, 1995). Changeux asserts that mathematical entities
‘exist in the neurons and synapses of the mathematician who produces them’
(ibid., p. 12), while Connes claims that he finds in the world of mathematics
‘a more stable reality than the material reality that surrounds us’ (ibid.). Two
radically different metaphysical positions stand opposed to each other in this
confrontation.
Metaphysics
One of the fundamental questions in philosophy is how we should conceive
that epistemology (knowledge) and ontology (being) are related to each other.
At one extreme, someone like Immanuel Kant divorced the two. In his opinion,
all we can know are phenomena, the appearances of things, while noumena,
things in themselves, are hidden from our view. At the other extreme stands the
realist, for whom epistemology and ontology are closely correlated and what
we know should be taken as a reliable guide to what is the case. Almost all sci-
entists, wittingly or unwittingly, are realists. If they did not believe that scien-
tific knowledge is telling us what the world is actually like, it is difficult to see
what would justify the great expenditure of labour and mental energy involved
in pure scientific research. Even in science, however, physics constrains but
does not fully determine metaphysics. There is no simple entailment between
the two. Their mutual relationship is similar to that between the foundations
of a house and the variety of edifices that might eventually be erected upon
them. For example, quantum physics is undoubtedly intrinsically probabilistic,
but does this uncertainty arise from a necessary ignorance of deterministic fine
detail or is it the sign of an actual intrinsic indeterminacy present in nature?
While most physicists follow Niels Bohr and his successors in taking the latter
view, David Bohm showed that there is an alternative interpretation of the
theory that yields identical physical predictions but which corresponds to the
former option of a veiled determinism (Bohm and Hiley, 1993). Since there
is no difference in the empirical consequences involved between the two, the
choice between Bohr and Bohm cannot be made on strictly physical grounds
but appeal must be made to metaphysical considerations, such as naturalness
and lack of contrivance.
Rather similar problems arise in relation to the nature of mathematical
entities. There is a vast and impressive body of mathematical knowledge.
What, if any, kind of reality is accessed by this knowledge? In the course of
defending his position, Changeux says that he adopts ‘a naturalist position that
makes no reference whatsoever to metaphysical assumptions’ (Changeux and
Connes, 1995, p. 213). How difficult it is to see ourselves as others see us!
Anyone who outlines a world view circumscribing the scope of reality, whether
it is narrowly or capaciously conceived, is making a metaphysical assertion as
surely as they are using prose to express their convictions. The materialist is no
exception to this rule. It is a common illusion, often entertained by those taking
a reductionally physicalist position, to suppose that somehow they are exempt
from having to make prior metaphysical assumptions, other than a reliance on
the reliability of science. Yet the fact is that science has purchased its very
great success by the limited character of its actual aims. We have seen from
the example of quantum physics that its discoveries constrain metaphysical
thinking, but they are by no means sufficient in themselves to determine what
its conclusions should be. The reduction of thought solely to physical states
of neural networks is not a deduction from neurobiology, but a metaphysical
M AT H E M AT I C A L REALITY 29
assumption imposed upon that scientific discipline. Of course, I do not chal-
lenge the belief that there is a connection between the workings of the mind
and the behaviour of the brain—I certainly accept that human beings are
psychosomatic entities—but the nature of that relationship is a not a question
that can be settled by neurological investigation alone, however important and
interesting that investigation undoubtedly is. Metaphysical questions demand
metaphysical answers, which have to be supported by metaphysical arguments.
Contemporary society, in striking contrast to the thinking of many previ-
ous ages, seems to treat materialism as the natural default position, scarcely
requiring any argument in its defence. Yet the world picture it presents is
that of a kind of lunar landscape, with complex, replicating and information
processing systems as its inhabitants, but with no persons in it. Much of what
makes human life valuable and satisfying is dismissed as epiphenomenal. No
due acknowledgement is given of the creative powers of imagination involved
in the intellectual enquiry that gave birth to science and mathematics. Personal
experience, which is the foundation of all our most significant encounters with
reality, rather than being accorded the privilege it deserves, is dismissed with
an unwarranted suspicion of its importance. Our mental life—the actual source
of all our knowledge—is treated as if it were a by-product of the material, in a
curious replacement of the direct by the abstracted.
In the materialist perspective, human beings are simply seen as computers
made of meat. The unsatisfactoriness of this as far as mathematical experience
is concerned seems clear. Mathematical thinking is more than computational
efficiency; mathematical insight is not confined within the Gödelian limits of
finitely axiomatized systems, a point that has been emphasized particularly by
Roger Penrose (Penrose, 1989).
Mathematical reality
Considerations relating to the issue of the reality of a noetic world of math-
ematical entities bear some analogy to similar arguments that can be made
defending the reality of the physical world against the critiques of the idealists.
However, before going on to consider these metaphysical arguments, one must
first be clear what results might be expected of them. The character of the
conclusions reached will be insightful and persuasive, rather than logically
coercive. The strict language of ‘proof’, with the implication that only a fool
could disagree, is inappropriate in this field of discourse. No one can force an
intransigent sceptic to give up their position, however arid and implausible it
may be. The solipsist, and the person who maintains that the world and our
memories of it came into being five minutes ago, are both logically invulnera-
ble in their absurdities. The best that can emerge from metaphysical disputation
is an argued claim to have attained the best explanation that is available.
The first of the analogies between human encounters with the physical
world and with the mathematical world relates to the consistency of perception,
and the mutual coherence of account, reported by different observers. Connes
summarizes this argument by saying,
What proves [too strong a word!] the reality of the material world, apart from
our brain’s perception of it? Chiefly the coherence of our perceptions and
their permanence. . . . And so it is with mathematical reality: a calculation
carried out in several different ways gives the same result, whether it is done
by one person or by several.
Changeux and Connes (1995, p. 22).
A second argument appeals to the richness to be found in independent real-

ity. Science’s exploration of the physical universe is the story of the discovery
of an apparently unending depth of rational structure and relationality, unveiled
as level after level of the world is disclosed to human enquiry. The richness thus
revealed is strongly persuasive that its source lies outside the limited human
mind of the investigator. For mathematics, Connes invokes Gödel’s theorem,
with its implication that the richness of arithmetic will never be contained
within a system of finite axiomatization, to make an analogous point about
mathematical reality. He says that the theorem means that ‘the quantity of
information contained in the set of all true propositions about the positive inte-
gers is infinite’, going on to comment ‘I ask you: isn’t that the distinguishing
feature of a reality independent of all human creation?’ (ibid., p. 160).
A third argument, related to the last, points to the element of surprise
involved in the exploration of an independent reality. A powerful support for a
realist interpretation of physical science can be found in the manner in which
the universe frequently resists prior expectations about its character, forcing
upon the physicist concepts that would never have been accessed without the
relentless pressure exerted by the stubborn nudge of nature. The resulting
feel of the scientific endeavour is that of discovery rather than construction.
The counterintuitive ideas of quantum physics are perhaps the most striking
example of this phenomenon. Who would have supposed that the apparent
ambiguity of wave/particle duality was a rational possibility without being
driven to it by the stubborn facts of the observed character of light? In a
somewhat similar way, totally unanticipated riches are revealed to the explorers
of the mathematical world. Connes’ favourite example is provided by the 26
‘sporadic’ finite simple groups, which defy classificatory incorporation into
such general categories as cyclic groups of prime order (du Sautoy, 2008).
A more pictorially accessible example would be the endlessly proliferating
structure of the Mandelbrot set, deriving from a deceptively simple-looking
and concise definition.
Considerations of this kind help to explain the conviction held by many
mathematicians that they are engaged in the discovery of actually existing
entities and their properties, and not merely the invention of pleasing intel-
lectual games, indulged in simply to exhibit their skill. In his book, A Math-
ematician’s Apology, the distinguished mathematical analyst G. H. Hardy
stated his conviction that ‘mathematical reality lies outside of us, that our
function is to discover or observe it, and that the theorems which we describe
grandiloquently as our “creations”, are simply our notes of our observations’
(Hardy, 1940: 1967, pp. 123–124). Of course, our material brains are involved
as instruments in making these observations, just as they are in our making
observations of the physical world around us, but in neither case should
the means of perception be equated with the realities perceived. Changeux’s
attempt to reduce mathematical entities to items of synaptic storage is to be
resisted as a category mistake, as crass as identifying literature with the ink
and paper by means of which it is recorded.
The plausibility of the concept that mathematical research is an act of
noetic exploration is strengthened by the role played in mathematical thinking
by intuitive perception and unconscious creative activity. Something is going
on that seems much more profound than can be described by a banal concept
of computational processing. There are well-documented cases of discovery in
which, after intense conscious engagement has failed to yield the solution of a
deep problem, a period of fallow disengagement is followed by a moment of
illumination in which the answer emerges into consciousness, essentially fully-
formed and only needing extended technical labour to complete the details
of proof. A well-known case involved the nineteenth-century mathematician,
Henri Poincaré. He had been wrestling with a problem connected with the
theory of Fuchsian functions, but he had made no progress. Consequently,
Poincaré decided to give it a rest and to take a holiday. At the very moment
of his departure, the complete solution to the problem sprang into his mind
unbidden. So sure was he that he had made the breakthrough, that he continued
on holiday, only engaging with the technical mopping-up operation on his
return. Perhaps the most striking example of the existence of profound intu-
itive mathematical powers was given by Hardy’s Indian colleague, Srinivasa
Ramanujan. This self-taught genius displayed an astonishing ability to write
down deep theorems in number theory that he had discovered, not by explicit
rational argument but by a tacit process of intuitive encounter. It is surely more
persuasive to understand Ramanujan’s great gifts as the consequence of an
ability to access and explore an existing noetic world, rather than there being
simply some fortuitous tricks of his neural organisation.
Evolution
A final argument for taking the independent reality of mathematical entities
seriously derives from asking how it might have been that profound mathe-
matical ability arose in the course of hominid evolution. It seems clear enough
that some very modest degree of elementary mathematical understanding—
the ability to count, simple notions of Euclidean geometry, and the capac-
ity to make simple logical associations—would have provided our ances-
tors with valuable evolutionary advantage. But whence has come the human
capacity to go far beyond matters of everyday utility, to attain the ability
to conjecture and eventually prove Fermat’s Last Theorem, or to discover
non-commutative geometry? Not only do these powers appear to convey
no direct survival advantage, but they also seem vastly to exceed anything
that might plausibly be considered a fortunate spin-off from such mundane
necessity.
The power of evolutionary explanation depends critically on getting the
environmental factors right, as much as it does on getting the genetic factors
right. If the context within which hominid evolution took place was solely
that of the physico-biological dimensions of reality, as strict neo-Darwinian
orthodoxy supposes, the coming-to-be of human mathematical ability would
seem to be an inexplicable excess. Yet one can take Darwinian explanation
absolutely seriously without having to suppose it to be a totally adequate
account of absolutely everything that has happened. If mathematical entities
constitute an independent realm of reality, then mathematics has always been
‘there’, even before mathematicians emerged. It formed the noetic context
within which that emergence eventually took place. While survival pressures
would favour the initial development of a brain structure that afforded access
to limited arithmetical and geometrical thinking, once that modest degree of
contact had been established with mathematical reality, further new devel-
opmental factors would come into play. The drive to assist physical survival
would be supplemented by the effects of a mental influence that one may
call ‘satisfaction’ (Polkinghorne, 2005, pp. 54–55). Intellectual delight would
then draw our ancestors into an exploration of the noetic world of math-
ematical entities, beguiling them to progress far beyond the modest needs
of everyday practicality. Doubtless the development of mental perceptive
power that this involved was made possible by the epigenetic plasticity of
the human brain, much of whose complex structure derives not from genetic
inheritance, but from response to the shaping influence of experience. Belief
in the reality of mathematics makes intelligible our human ability to be
mathematicians, a capacity that otherwise would have seemed inexplicably
gratuitous.
Unreasonable effectiveness
If mathematical entities are a part of reality, then one might expect that the
ontological realm of their existence is not an isolated domain, disconnected
from all else, but that it has subtle connections with other dimensions of the
real. A very striking example of this happening is provided by the connection
found to exist between theoretical understanding in physics and mathematical
properties. It is an actual technique for discovery in fundamental physics to
seek theories that are formulated in terms of equations possessing the unmis-
takable character of mathematical beauty. This beauty is a rather rarefied form
of aesthetic experience, but it is one that mathematicians can readily recognize
and agree about. It involves qualities such as elegance and economy and the
property of being ‘deep’; that is to say, extensive and surprisingly fruitful
consequences are found to derive from an apparently simple starting point.
The physicists’ search for beautiful equations is no mere aesthetic indulgence
but a heuristic strategy which has proved its worth time and again in the three-
century history of modern physics. Paul Dirac, one of the founding figures
of quantum theory, made his remarkable discoveries through a lifelong and
highly successful pursuit of mathematical beauty. He once said that it is more
important to have beauty in your equations than to have them fit experiment!
Of course, Dirac did not mean that empirical adequacy was ultimately dispens-
able. No scientist could think that. If you have solved the equations of your new
theory and found that the answers do not appear to agree with experiment, that
is undoubtedly a setback. However, it is not necessarily absolutely fatal. No
doubt you have had to have recourse to some approximation scheme in getting
your solution, and maybe you have just made an inappropriate approximation.
Or maybe the experiments were wrong—we have known that happen more
than once in physics. So there would still be at least a residue of hope. But
if your equations were ugly. . . well, there was no hope. The whole history of
physics was against you.
Dirac’s brother-in-law, Eugene Wigner, who also won a Nobel Prize for
physics, once called this remarkable ability of mathematical beauty to unlock
the secrets of the physical universe its ‘unreasonable effectiveness’. How does
it come about that this apparently abstract subject can illuminate our under-
standing of the structure of the physical world? Why are the beautiful patterns
of pure mathematics, discovered by the mathematicians in their studies, so
often found actually to occur in the structure of the world about us? This is not
the place to pursue that particular issue in detail (I personally look to natural
theology for an answer (Polkinghorne, 1998, ch. 1)). It is sufficient for our
present purpose simply to note the fact, and its implication of a deep mutual
entanglement of the physical and the mathematical. Few doubt the reality of the
physical world; they should be prepared to consider acknowledging a similar
reality of the mathematical world that intertwines with it.
Mathematics also entangles with other dimensions of reality. I wish to take
very seriously human encounter with the realm of beauty. I do not think that
our aesthetic experiences are simply some kind of epiphenomenal froth on top
of what basically is just a physical substrate, but they are a form of access
to yet another dimension of reality. Of course, music involves vibrations in
the air, but its appreciation is not to be reduced to the fourier analysis of
those vibrations. There is a deep mystery about the way that the impact of
packets of sound waves on the eardrum can evoke in us what I believe to be
the valid experience of encounter with a timeless beauty. There is an often
recognised kinship between mathematics and music, expressed not only by the
way that individuals frequently display interests and skills in both, but also by
the patterns that both are found to share, particularly in the case of contrapuntal
music.
Another aesthetic experience in which patterns play a vital is exhibited
to us by Islamic art. Marcus du Sautoy (2008, ch.3) has given a fascinating
discussion of the symmetries that underlie the elaborate decoration of the walls
of the Alhambra Palace in Granada. In the nineteenth century, group theorists
were able to show that there are just 17 different basic kinds of symmetry that
can be present in such regular coverings of the plane. All 17 are present in the
decorations of the Alhambra, which was built in the thirteenth and fourteenth
centuries. The Islamic artists involved did not know the group theory, but they
had intuitive access to the mathematical reality that their work expressed.
Conclusion
The criterion for assessing the persuasiveness of a metaphysical position is the
seriousness with which it treats, and the adequacy with which it can contain,
the great swathe of basic human experience which it is seeking to make
intelligible through its insight. Schemes that are made parsimonious simply by
the illegitimate Procrustean strategy of the excision of what, from their point
of view, is inconvenient to take into account, are certainly to be rejected. The
argument of this chapter has sought to show that an approach that seeks to take
the actual character and achievements of mathematics with the due seriousness
that they deserve, is one that is best formulated in a metaphysical context that
acknowledges the reality of a noetic world of mathematical entities.
Comment on John Polkinghorne’s
‘Mathematical reality’
Mary Leng
Attacking the thorny issue of ‘creation vs. discovery’ in mathematics, John

Polkinghorne’s contribution to this volume argues that mathematicians are
engaged in inquiry into the nature of a ‘noetic realm’ of mathematical entities,
and therefore in discovery, rather than creation, of their mathematical sub-
ject matter. Polkinghorne contrasts this picture with a physicalist alternative,
according to which human mathematical activity would have to be explained
without reference to such a realm of mathematical objects, for example, as the
construction of ‘ingenious intellectual puzzles’ or ‘the painstaking unravelling
of a monstrous logical tautology’.
Polkinghorne quite rightly notes that no deductive proof can be provided
for his, or indeed any, account of the fundamental nature of mathematics: our
philosophical theory here is underdetermined by the empirical evidence, so
at best inductive reasons (reasons which do not establish, but at best make
probable, their conclusion) can be given to prefer Polkinghorne’s noetic realm
hypothesis to the hypothesis of physicalism. Polkinghorne further notes that
physicalism should be viewed as just one metaphysical hypothesis among
others, rather than as a default position to be held on to unless one is provided
with conclusive reasons for its rejection. Polkinghorne’s strategy is therefore
to consider the noetic realm hypothesis alongside physicalism’s denial of this
realm, subjecting both to the evidence provided by a variety of phenomena,
and to argue that these phenomena speak against physicalism and in favour of
the existence of a noetic realm of mathematical objects.
That a question is only amenable to inductive considerations, and not
deductive proof, does not of course take away from the meaningfulness,
value, importance, or indeed tractability of that question. Indeed, while pure
mathematics thrives on deduction, inductive reasoning is the lifeblood of the
empirical sciences, where almost every important theoretical question requires
a leap beyond what can be narrowly deduced from empirical observation.
Polkinghorne’s discussion makes use of two distinctive types of inductive
reasoning that are commonplace in empirical science in order to argue for
his metaphysical conclusion: arguments by analogy, and inferences to the best
explanation.
Arguments by analogy take the general form: X is like Y in respects
a, b, . . . . Therefore (probably), X is also like Y in respect z. This form of reason-
ing is essential, for example, in drawing conclusions about real situations from
models of those situations, where the inferences can be quite mundane. (The
city is like our map with respect to the placement of the landmarks, streets,
and relative distances that we can see. Therefore (probably), it is also like our
map with respect to the location of the town hall. So to get to the town hall we
should go straight ahead and take the third street on the left.) More creatively,
Newton made use of an analogy between planets and balls to argue that the
motion of a planet around the Sun should be relevantly similar to the motion
of a ball being swung around on an elastic string (with the circular motion
being produced by the ‘pulling’ force exerted by the elastic in the ball case,
and by the Sun in the planet case). Speculative analogies of this sort require
some crafting. For any values of X and Y, we will be able to find some respects
in which they are similar. The trick is to find enough relevant similarities to
warrant the inference to a further similarity.
Polkinghorne’s argument by analogy (which he draws from Alain Connes’s
response to Jean-Pierre Changeux’s radical materialist position) makes use of
three respects in which mathematical inquiry is like inquiry into the material
world. His conclusion is that mathematical inquiry is also like inquiry into the
material world in having an independent, objective, real world of objects as
its subject matter. Polkinghorne himself suggests that each of these respects
is, by itself, reason enough to draw his realist conclusion, but taken together,
they can only strengthen the overall analogy. In assessing this analogy, then,
it is important to consider whether the respects in which the two forms of
inquiry are similar (coherence of ‘perceptions’ across time and observers;
richness; and capacity to surprise) are relevant in relation to the further sim-
ilarity Polkinghorne wishes to claim (i.e., concerning an independent realm
of objects). Polkinghorne has certainly picked on markers of objectivity in
mathematical reasoning: that different reasoners using different methods in
isolation from one another can still agree on mathematical conclusions, for
example, certainly suggests that mathematicians are not free to draw whatever
conclusions they please. But are the similarities relevant to Polkinghorne’s
further claim, that mathematical inquiry, like physical inquiry, concerns an
objective, independent, realm of objects?
In answering this question, one might consider whether an alternative
account of the nature of mathematics can explain the phenomena Polking-
horne mentions as well as, or better than, does Polkinghorne’s noetic realm
hypothesis. If some alternative, and plausible, explanation of the similarities
between mathematical and physical inquiry can be provided, which does not
appeal to a noetic realm, then the strength of the original argument from
M AT H E M AT I C A L R E A L I T Y: C O M M E N T 37
analogy is undermined. This brings us, then, to the second of Polkinghorne’s
argumentative strategies: inference to the best explanation (IBE).
Our late colleague, Peter Lipton, describes this form of inference as
follows:
Given our data and our background beliefs, we infer what would, if true,
provide the best of the competing explanations we can generate of those
data (so long as the best is good enough for us to make any inference
at all).
Lipton (1991: 2004, p. 56)
As Lipton points out, the word ‘best’ needs some clarification here. In
particular, we can distinguish between
the explanation best supported by the evidence, and the explanation that
would provide the most understanding or, in short, between the likeliest and
the loveliest explanation.
Lipton (1991: 2004, p. 59)
Advocating inference to the likeliest explanation is, as Lipton points out, rela-
tively uncontroversial but, sadly, fairly unhelpful—if we had a way of knowing
what scenario was most probable, we would certainly infer that over the alter-
natives, but IBE is surely intended in part as a method for discovering which of
alternative possibilities is more probable. For inference to the best explanation
to be a practical rule of theory choice, we need to build on our account of what
makes an explanation lovely in the sense of providing understanding—perhaps
by invoking theoretical virtues such as simplicity, non-ad-hocness, unifying
power, and so on. Whatever our account of ‘loveliest’ amounts to, Polking-
horne clearly thinks that the noetic realm hypothesis provides the loveliest
explanation of the similarities between mathematical and physical inquiry he
indicates. Lovelier, certainly, than Changeux’s account of mathematical objects
as existing ‘in the neurons and synapses of the mathematician who produces
them’—could those neurons and synapses really contain the rich, surprising,
and universally accessible subject matter of mathematical inquiry? And there
are further phenomena for which Polkinghorne thinks the noetic realm also
provides the loveliest explanation. In particular, Polkinghorne points to cases
of sudden and deep mathematical insights (such as those reported by Poincaré
and Ramanujan); the capacity of human reasoners to go beyond the narrow
range of mathematics that we could expect to give us evolutionary advantage;
and the ‘unreasonable effectiveness’ of mathematics in empirical science. To
take just the last of these, Polkinghorne argues that seeing mathematics as
one dimension of reality renders its effectiveness in finding out about physical
reality unsurprising, since one should expect the mathematical realm to ‘have
subtle connections with other dimensions of the real’. But that two systems of
objects exist cannot by itself be enough to explain why facts about one system
are relevant in finding out about facts about the other: my kitchen exists, and so
does the solar system, but if it turned out I could reliably divine facts about the
solar system by reasoning about the contents of my kitchen, one might still find
the effectiveness of this reasoning unreasonable. More needs to be said about
why the mere existence of a mathematical realm should render its effectiveness
reasonable.
There are those who look with suspicion on metaphysical speculation,
on the grounds that it is often unclear what kind of evidence could count
in favour of, or against, any particular metaphysical hypothesis. One of the
great merits of Polkinghorne’s contribution to this volume is that it lays out
in clear and precise terms what would be required of his opponent in this
debate: find a better explanation of the phenomena in question that does not
make use of the noetic realm hypothesis. For those (myself included) who
think alternative explanations can be found, Polkinghorne’s paper presents a
formidable challenge.
Reply to Mary Leng
John Polkinghorne
I am grateful to Mary Leng for her helpful analysis of the arguments I sought
to deploy in defense of mathematical realism. Of course she is right that the
existence of two objects does not itself imply a mutual connection, but the rela-
tionship between mathematics and physics is a deep and apparently intrinsic
(unlike her table and the solar system), and I still maintain this encourages the
thought that each is a part of a greater reality.
4
Mathematics, the mind, and
the physical world
Roger Penrose
Does mathematics have an independent reality? Or is it simply a product of

human thought and culture? Or perhaps it is merely an idealization, abstracted
from what we find to be a mathematical organization that is well approximated
in the structure and dynamical behaviour of the physical world?
In this chapter I shall attempt to address these two aspects of mathematical
Platonism: the issue of the independent reality of mathematics and the separate
issue of whether there is a fundamental dependence of physical behaviour on
such a pre-existing mathematics. These two themes, which perhaps are often
confused with one another in the minds of those who are uncomfortable with
mathematical Platonism, form the essential subjects of this chapter. I shall try
to illustrate my own position on these issues in relation to Fig. 4.1, where
I have schematically depicted1 three ‘worlds’: the physical, the mental and
the mathematical, together with what I regard as the three deeply mysterious
connections between them.
The first aspect of mathematical Platonism referred to above is the nature
of ‘Mystery 0’, namely whether the ‘world of mathematics’ arises merely as
a product of our mental activities, having no reality beyond this, or whether
it is to be assigned an independent existence of its own. And if the latter,
whether or not we, in principle, have access to this world in its entirety. The
second, separate, issue—depicted as ‘Mystery 1’—has to do with the role
of mathematics in physical theory. Does the undoubted utility of mathemat-
ics in our understanding of the physical world reflect merely our facility in
organizing observational data into some comprehensible form, where those
aspects of physics which work well as mathematical theories are assigned an
undue importance, merely because they do work well? Or is it really true, as
1 This figure first appeared in Penrose (1994), but I have used it frequently elsewhere,
such as in Penrose (2004).
Platonic
mathematical
world
Mental
world
1
Physical
world
Fig. 4.1 The three ‘worlds’: the physical, the mental and the mathematical.
many theoretical physicists appear to believe, that there is a deep and precise
underlying dependence of the operation of the physical world on a pre-existing
mathematical order—a mathematical order that appears to have great beauty
and sophistication—which is there to be discovered and not simply imposed
upon Nature as a feature of our gropings towards understanding?
To complete the triad of mysteries connecting these worlds is ‘Mystery 2’,
which is concerned with the relation between physical reality and mentality,
particularly conscious mentality. How does consciousness arise in a world
which seems to be governed by entirely impersonal mathematical operations?
Or is consciousness in some sense primary, its presence being an essential
prerequisite for the very existence of a structure that we could, in any sense,
call a ‘universe’? Is it merely the complication, or perhaps some other sophis-
ticated quality, in the construction of our brains that allows this mysterious
phenomenon of consciousness to come about? And, if so, is this complication
or sophistication to be understood solely in computational terms, as seems to
be a common viewpoint in our present computer age. Or is there some other
essential prerequisite to consciousness, that cannot be understood in purely
computational terms? If the latter, is this ingredient something lying hidden in
M AT H E M AT I C S , THE MIND, AND THE PHYSICAL WORLD 43
the physics that we presently use for our descriptions of the operations of the
world? Or must we search for deeper (mathematical?) theories for a physical
description of consciousness to be possible. Or might we perhaps have to look
even farther afield, to an understanding that lies essentially beyond any kind
of science whatever, as could be the implication of an essentially religious
perspective on these issues?
In this chapter I shall be concerned mainly with Mysteries 0 and 1, as
the topic of the Symposium had to do specifically with mathematics. But it
is my opinion that an adequate discussion of these two mysteries cannot be
completely divorced from some discussion of Mystery 2. I shall argue the
case (by appealing to Gödel incompleteness) that the very fact that our minds
are capable of comprehending sophisticated mathematical arguments—at least
under favourable circumstances—leads us to the conclusion that the operation
of conscious minds cannot be entirely computational and, accordingly, that our
minds cannot be the product of an entirely computational physics. It does not
appear to be the case that the physical laws that we presently understand con-
tain anything that is essentially non-computational (where merely ‘random’ is
not to be considered as ‘essentially non-computational’). The conclusion from
this is that there must be something beyond our present-day physical laws that
is operative in the actions of a conscious mind. It is my own viewpoint that this
is strongly indicative of the actions of a conscious brain being dependent upon
areas of physics (probably at the quantum/classical boundary) that lie outside
the scope of our present-day physical theories; yet that the needed physical
revolution may not itself lie too far beyond what is presently understood.
The issue of Mystery 0 is indeed a closely related matter. Part of the reason
for regarding our access to mathematical truth as being something ‘mysterious’
lies in the nature of our ability to perceive the truth of various particular math-
ematical assertions. As Gödel (and Turing) have demonstrated, if we accept
any particular computationally checkable system of procedures P as providing
valid methods of mathematical proof, then we must equally accept the truth
of some proposition G(P), where the truth of G(P) lies beyond the scope of
the procedures P. Consequently, our methods of ascertaining mathematical
truth cannot be entirely reduced to computational procedures that we accept
as valid. Although various logicians have attached different interpretations to
this conclusion, in my own view, it has the clear implication that there must
be something essential lying outside pure computation that is operative in
conscious understanding. (For further discussions of this point, see Penrose,
1997.) But what really is going on in the activity of a conscious mind when
it becomes convinced of the truth of some mathematical proposition remains
profoundly mysterious.
I maintain, also, that this is strongly indicative of mathematical truth being
something objective (as was, indeed, Gödel’s own view), and is not merely
some ‘game’ based on arbitrary rules that arise out of human culture. Yet,
I am prepared to accept that there could be ‘degrees of Platonism’, where
some mathematicians might regard the truth of some proposition P as being
an ‘objective’ matter, whereas another might take the view that the ‘truth’ or
‘falsity’ of P is a matter of opinion, depending upon what ‘man-made’ axiom
system is adopted. I believe that it is clear, from Gödel’s construction, that
certain areas of mathematics are indeed ‘objective’, and therefore have an
existence that is outside ourselves. Such an area would be the truth of what
are called ‘1 -sentences’, these being assertions of the form ‘such-and-such a
computational procedure never terminates’ (where ‘computational procedure’
means ‘Turing-machine action’). An example of a 1 -sentence is the famous
Fermat’s Last Theorem. It seems to me that the truth or falsity of a 1 -sentence
is something entirely objective, and so the truth values of 1 -sentences have
a Platonic reality which is not in doubt (although there may be an element of
subjectivity to the issue of whether or not some particular 1 -sentence has
actually been established).
On the other hand, the objectivity of more sophisticated assertions such
as the truth or otherwise of Cantor’s (generalized) continuum hypothesis is
perhaps more questionable, and it would seem to require a stronger form of
mathematical Platonism to require that all such assertions are true or false in an
absolute sense,2 and not merely dependent upon some particular ‘man-made’
axiom system. The issue of whether or not some particular mathematician is,
or is not, a Platonist usually refers to Platonism in this stronger sense. My
own position is not to be particularly troubled by this issue, as a relatively
weak form of Platonism seems to be perfectly adequate for most arguments of
relevance to physics.
Accepting that, indeed, we need to consider a ‘Platonic mathematical
world’ that is merely large enough to encompass the description of physical
laws, we have an additional case for its ‘existence’, lying beyond anything
that is merely brought into being by human culture or imagination. For the
operations of the physical world are now known to be in accord with ele-
gant mathematical theory to an enormous precision. One particularly striking
example (see, for example, Hartle, 2003) is the double-neutron-star system
PSR 1913+16, which has been under observation, now, for some 30 years,
and the agreement between observation of the timing of its pulsar signals
and Einstein’s general theory of relativity (to something like one hundredth
of a thousandth of a second over that entire period) is phenomenal. This
indicates an extraordinary concurrence between the workings of the natural
world at its most fundamental levels (here the very structure of space and
time) and sophisticated mathematical theory. It makes no sense to me that this
concurrence is merely the result of our trying to fit the observational facts
2 It should be made clear that the results of Gödel and Cohen, showing that the
continuum hypothesis is independent of the standard axioms of set theory, do not
in themselves answer the question of whether or not this hypothesis is true in some
absolute sense; see Cohen (1966).
M AT H E M AT I C S , THE MIND, AND THE PHYSICAL WORLD 45
into some organizational scheme that we can comprehend; the concurrence
between Nature and sophisticated beautiful mathematics is something that is
“out there” and has been so since times far earlier than the dawn of humanity,
or of any other conscious entities that could have inhabited the universe as we
know it.3
Acknowledgement
I am grateful to NSF for support under grant PHY 00-90091.
3 More details about the arguments given here are to be found in Penrose (2011).
Comment on Roger Penrose’s
‘Mathematics, the mind, and
the physical world’
Gödel’s theorems and Platonism
Michael Detlefsen
Roger Penrose’s chapter contains a number of claims and ideas that warrant,
and have received, extensive discussion. In this note, I will focus on the
following two claims that are central to his view of the significance of Gödel’s
theorems.
I: Gödel’s incompleteness theorem(s) demonstrate that ‘if we accept any
particular computationally checkable system of procedures P as providing
valid methods of mathematical proof, then we must equally accept the truth
of some proposition G(P), where the truth of G(P) lies beyond the scope
of the procedures of P.’
II: Claim I ‘has the clear implication that there must be something essen-
tial lying outside pure computation that is operative in conscious
understanding.’
We can restate Penrose’s claims more clearly and in more familiar
terminology as follows:
I∗ : For any formal system P, if we accept all of P’s axioms as true, and all its
rules of inference as valid, we are rationally obligated to accept P’s Gödel
sentence G(P) (and its P-equivalent consistency formulae Con(P)) as true
too.
II∗ : Claim I∗ clearly implies that there is a set A of sentences we are rationally
obliged to accept that cannot be formalized (i.e., is not a computably
enumerable set).
There is little reason I can see for confidence in either I∗ or II∗ . G(P) and
the usual consistency formulae Con(P) equivalent in P to it are not logically
M AT H E M AT I C S , T H E MIND, AND THE PHYSICAL WORLD: C O M M E N T 47
implied by P. This means that they logically imply sentences that are not
logically implied by P. There is generally speaking no reason to believe that
all reasoning which supports P will equally support these ‘extra’ implications.
Consequently, there is no general reason to think that rational acceptance of P
will rationally oblige acceptance of G(P)/Con(P).
This may seem wrong to those who believe that evidence capable of
justifying P must also be capable of justifying belief in its consistency. It is
important to realize, however, that to justify the consistency of P and to justify
G(P) or Con(P) are not the same thing. Justification that P is consistent is not
in and of itself justification of G(P) or Con(P). For the latter, justification for
the following supplementary proposition is also required.
Supp: If P is consistent, then G(P)/Con(P).
Such justification will not, however, necessarily be included in the evidence
one might have for P’s consistency.
This is not to deny, of course, that there may be justification for Supp. Nor
is it to deny that there may be rationally compelling grounds for P that include
rationally compelling grounds for G(P)/Con(P). It is intended only to indicate
the non-triviality of Supp and thus to demur to claims such as I∗ , which
suggest that rational acceptance of G(P)/Con(P) is somehow presupposed by
rational acceptance of P.
There are other reasons too to question Penrose’s argument, but there is no
space to enter into these here.
5
Mathematical understanding
Peter Lipton
I am alas not a philosopher of mathematics, but I am interested in the nature

of scientific explanation and I am hoping that we can have a discussion of the
nature of mathematical understanding, a topic on which some other members
of the Symposium are expert. It would be enlightening for me if we could
compare physical and mathematical explanation: the comments below flag
some natural focal points.
My own work on explanation is motivated in part by two simple and gen-
eral considerations. The first is that there is a gap between knowing that some
phenomenon occurs and understanding why it occurs. Knowing is generally
necessary but not sufficient for understanding. Thus most people know that the
same side of the Moon always faces the Earth, but few people understand why.
(The question begins to nag once you realize that the phenomenon requires that
the Moon’s period of orbit around the Earth be exactly the same as its period
of spin.) Much of the work on the philosophy of explanation can be seen as
attempts to answer the question of what bridges this gap between knowing
that and understanding why. And the existence of the gap places a helpful
constraint on adequate answers to the question of how to make the connection,
since it shows that any model of understanding that treats what is necessary
for mere knowledge as sufficient for understanding will be inadequate. Thus
the Hempelian idea that a good explanation provides understanding of the
phenomenon in question by providing reasons to believe that the phenomenon
was to be expected is inadequate, since reasons for belief are in these cases
required simply to know that the phenomenon occurs.
The second consideration that motivates me places an additional constraint
on an acceptable answer to the gap question. This is the ‘why-regress’, a
feature of explanation that many of us discovered as young children, to our
parents’ consternation. Given an acceptable answer to a ‘why-question’, it is
almost always possible to ask why about the answer itself. In my view, the
moral of this regress is not that explanation is impossible but rather that, in a
manner of speaking, we can explain with what we don’t understand. B may
explain A and hence provide understanding of why A is the case even if B is
not itself explained. Thus the fact that my student’s computer crashed explains
why her essay was late, even if nobody knows why the computer crashed.
What this shows is that understanding is not some kind of ‘super-knowledge’
that is transmitted from explanation to phenomenon explained. Understanding
seems not to be some special epistemic state, but rather additional information,
information that need not itself have a special epistemic status.
These two considerations—the gap between knowledge and understanding
and the why-regress—are uncontroversial in the context of physical explana-
tion; but perhaps not in mathematics. Thus one could deny that there is any
distinction between merely knowing that some mathematical statement is true
and understanding why it is true. This would be a close cousin of the denial that
there is a distinction between explanatory and non-explanatory proofs. And in
the mathematical case one could also resist the moral I drew from the why
regress, that the explanation need not have a special epistemic status. Thus one
might hold that mathematical understanding flows from the special epistemic
status of mathematical axioms, and that these axioms are where the regress
stops. Is mathematical explanation different from physical explanation with
respect to the gap and the regress?
Two of the most popular accounts of physical explanation are the causal
model and the necessity model. According to the former, to explain is to
provide information about the causal history of the phenomenon in question;
according to the latter it is to show that the phenomenon was in some sense
necessary—it had to happen. These two models seem strikingly inappropriate
for mathematical explanation. If one opted for the causal model, it looks
like there would be no mathematical explanations, since mathematical facts,
however one construes these, do not appear to enter into causal relations. So
no proofs would be explanatory. And if one opted for the necessity model, it
looks as though every proof would be explanatory. Indeed it looks as though
the contrast between knowing and understanding would disappear. For even
if one knows a mathematical truth not on the basis of any proof but rather on
the basis of expert testimony, one would still know that what one was told is
necessary, since we know in advance that, for any mathematical statement, if
it is true then it is necessarily true.
Fortunately, there are other models of explanation in stock. The obvious
one to reach for is some version of a unification model, according to which we
understand why something is the case when we see how it fits into a unified
pattern. This vague thought could be articulated in various ways, but at least
it seems to be a source or a criterion that might differentiate explanatory from
unexplanatory proofs. But a unification model may well leave out some sources
of mathematical understanding. Proofs by reductio might be good test cases.
It is natural to look here for example of non-explanatory proofs, but it is not
clear that these proofs fail to provide understanding (if indeed they do fail)
because they fail to unify. It may be more natural to say that they fail because
they do not show what ‘makes’ the theorem true, where ‘making’ is a form of
M AT H E M AT I C A L U N D E R S TA N D I N G 51
determination of which physical causation is only one species. So perhaps in
addition to a unificatory source of mathematical understanding, we ought to be
looking to articulate a notion of non-causal determination.
In these comments so far I have had in mind primarily the mathemati-
cal explanation of mathematical phenomena, but I would also like improve
my grip on apparently mathematical explanations of physical phenomena.
For example, suppose that parents notice that when their child is behaving
exceptionally badly, punishment is usually followed by some improvement
in behaviour, whereas when their child is behaving exceptionally well and is
rewarded, the subsequent behaviour is often somewhat worse. From this they
infer that punishment is more effective than reward. This inference is probably
unwarranted, because these are the patterns of behaviour one would observe
if both punishment and reward were entirely inefficacious: the pattern can be
explained simply by appeal to regression to the mean. Here we seem to have a
lovely explanation of a physical phenomenon in terms of a mathematical fact.
Here is another example. You throw a bunch of sticks into the air with a lot
of ‘English’, so that they tumble and spin as they fall. Now freeze the scene just
before the lowest stick touches the ground. What you find is that appreciably
more of the sticks are closer to the horizontal than to the vertical. Why is this?
It is because of the broadly mathematical fact that there are more ways for a
stick to be near the horizontal than there are for it be near the vertical. (Think
of a single stick with a fixed midpoint: there are only two ways for it to be
vertical, but indefinitely many ways for it to be horizontal. The asymmetry is
preserved for near the vertical and near the horizontal.)
These cases of mathematical explanations of physical phenomena raise a
number of interesting questions. Are they really non-causal explanations? Is
the mathematical fact really doing the explaining? What if anything do these
apparent cases of the mathematical explanation of physical phenomena tell us
about the mathematical explanation of mathematical phenomena?
So far I have written (and until recently I have thought) that understanding
is simply the flip side of explanation: ‘understanding’ is just the name for
what you get from an explanation. But I have begun to think that this is too
restrictive a notion of understanding, that while explanation is one route to
understanding, it is not the only one. The more radical form of this thought
is that there are forms of understanding that explanations do not provide. The
less radical form is that the same forms of understanding that explanations
provide may also be acquired by other routes. While not wishing to extend the
notion of understanding so far as to deprive it of any interesting content, I am
attracted to both forms. Thus working with a scientific theory may provide a
kind of intellectual know-how that is tantamount to a form of understanding
the phenomena, but one that is distinct from the kinds of understanding that
explanations provide. As for the less radical thought, it looks like most of the
characteristic cognitive benefits of explanation can be acquired by other means.
Take for example the necessitarian idea, the idea that an explanation may
provide understanding by showing that the phenomenon had to occur. It seems
that it is sometimes possible to show necessity and so to provide this kind of
understanding without explaining. Thus we come to understand why gravita-
tional acceleration is independent of mass by appreciating Galileo’s wonderful
thought experiment. Suppose that heavier things accelerate faster than light
things, and consider a heavy and a light mass connected by a rope. Considered
as two masses, the lighter one should slow down the heavier one, so the system
should accelerate slower than the heavier mass alone. But considered as one
mass, the system is of course heavier than the heavy mass alone and so should
accelerate faster than the heavier mass alone. But the system can’t accelerate
both slower and faster, so acceleration must be independent of mass.
Because it is a reductio, the thought experiment does not itself seem to
be an explanation. Nor does it appear that the route to understanding passes
subsequently through an explanation that the thought experiment supports.
Having absorbed Galileo’s argument, I understand why acceleration must be
independent of mass; but if you were to ask me to explain why acceleration
is independent of mass, I cannot do it: the best that I can do would be to give
you the thought experiment so that you can see for yourself. If, in the case
of physical phenomena, we can acquire understanding without explanation, is
the same true for mathematical understanding? This seems plausible, though
the Galilean example may not make the case, since while in that physical
instance coming to see the necessity may take one across the bridge from
knowing to understanding, as I have suggested above this may not be so in
purely mathematical instances, since here the necessity is already a part of
mere knowledge.
Although I have already raised more issues about mathematical under-
standing than we will probably have time or inclination to pursue in the
Symposium, I want to end by mentioning two more topics from my own work
that might suggest points of comparison between physical and mathematical
explanation. The first of these concerns the interest-relativity of explanation.
It is a familiar and plausible claim that what counts as a good explanation is
not determined by the phenomenon in question alone: it also depends on the
interests and background knowledge of the person asking the question. For
example, a good answer normally needs to give the inquirer something she
doesn’t already know.
In the context of a causal model of the explanation of physical phenomena,
the need to take interest into account is clear in light of the density of causal
histories. Behind every phenomenon there are innumerable causes, not all of
which are explanatory. Thus while that computer crash does explain why my
student’s essay was late, the Big Bang doesn’t, even though it is part of the
causal history of every event. Moreover, the same cause may be explanatory
for one person but not for another. Thus to take an example familiar from the
literature, given that the only cause of paresis is untreated syphilis, but most
people with untreated syphilis do not go on to contract paresis, one person
may find it explanatory to be told that Smith contracted paresis because he had
untreated syphilis while another person may rightly reject the explanation.
M AT H E M AT I C A L U N D E R S TA N D I N G 53
Some aspects of interest-relativity can be naturally analysed by giving
more structure to the why-question. For many why-questions do not take the
simple form ‘Why P?’ but rather have the contrastive form ‘Why P rather than
Q?’ The choice of the foil Q makes a difference, and people with different
interests may choose different foils. Thus if the real question is why did Smith
rather than Jones contract paresis, where Jones did not have syphilis, then
citing Smith’s syphilis will be explanatory, but if the question is why Smith
rather than Doe contracted paresis, where Doe also had syphilis, then it will not
be. (‘But Doe had syphilis too’, your interlocutor will reply.) In my view, many
of these contrastive questions generate a kind of triangulation that marks a
distinction between explanatory and unexplanatory causes. Roughly speaking,
what we need for explanation in these cases is a cause of P that ‘makes a
difference’ between P and Q, and this can be seen as a cause of P for which
there is no corresponding event in the case of Q. Smith’s syphilis explains why
he rather than Jones had paresis but not why he rather than Doe had paresis,
because Jones did not have syphilis while Doe did. Does any of this carry over
to mathematical explanations? Presumably here too there are various forms of
interest-relativity. But is there a place for a contrastive analysis? And if so,
how does the choice of contrast help to distinguish between explanatory and
non-explanatory information?
Finally, I cannot forbear mentioning my interest in inference to the best
explanation. This is the idea that scientists (and laypeople) often seem to use
explanatory considerations as a guide to inference. They infer that a hypothesis
is correct because although it is of course not the only hypothesis logically
compatible with the evidence, if it were true then it would provide the best
explanation of that evidence. If we want to articulate this thought, then one of
the things we need to do is to say more about what ‘best’ is supposed to mean
in the slogan ‘inference to the best explanation’. For example, should we take
‘best’ explanation to mean the likeliest, that is the most probable explanation,
or should we rather take it to mean the explanation that would if correct provide
the greatest degree of understanding, that is the ‘loveliest’ explanation?
Likeliest might seem the obvious choice, since we want what we infer
to have high probability. But I think this is the wrong choice in this context,
since it would make the explanationist idea almost vacuous, reducing it to the
claim that scientists infer what they take to be the most probable hypothesis.
The initial attraction of the idea of inference to the best explanation was that
it would illuminate scientists’ inferential practices, but to say that scientists
prefer what they take to be the most probable hypothesis is disappointing
on this score. We get a much more interesting account of inference if we
choose ‘loveliest’. It is far from trivial to say that scientists tend to infer that
the explanation that would if correct provide the greatest understanding is the
explanation that is also likeliest to be correct: far from trivial, and maybe even
approximately true. Those who find this line of thought attractive have their
work cut out for them, since now they need to say what makes one potential
explanation lovelier than another.
Does the idea of inference to the best explanation gain any purchase
in the context of mathematical reasoning? This might seem unlikely, since
inference to the best explanation is meant to provide a partial account of
non-demonstrative reasoning, whereas at least the paradigm of mathematical
reasoning is deductive. Who needs to appeal to the wishy-washy notion of
inference to the best explanation when you have a proof? But things look
different if we look at the context of discovery rather than of justification. For
the heuristic of mathematical investigation is no demonstrative process, and it
is possible that something like the idea of inference to the best explanation is
applicable here. This is a question we might be able to answer, if only we can
get a bit clearer about the nature of mathematical understanding.
Addendum on Peter Lipton’s
‘Mathematical understanding’
Stewart Shapiro
At the Castel Gandolfo Symposium, we had a most fruitful discussion of

Peter Lipton’s contribution, ‘Mathematical understanding’, in light of the other
papers presented at the Symposium, and other work by him and the partici-
pants. His keen understanding and wit, together with his modest, disarming
style, made the session especially interesting and compelling. With Lipton’s
sudden, shocking, and tragic death, there can be no update or revisions to his
chapter. I hope here to provide some background to the work, and to locate it
in some wider contexts.
Lipton was a philosopher of science, with special interests in the notions
of explanation and understanding. His book, Inference to the Best Explanation
(Lipton, 1991), is now close to being a classic—certainly required reading for
anyone with even a passing interest in the topic. The second edition appeared
in 2004. The focus of the book, and of much of his work, is on scientific expla-
nation and understanding. In a preliminary, programmatic way, his Symposium
contribution attempted to extend the concepts and ideas to mathematics.
As Lipton notes in his chapter, there is a clear and intuitive difference
between knowing that a given proposition is true or, as he puts it, knowing
that a given phenomenon occurs, and understanding why the proposition is
true, or why the phenomenon occurs. Presumably, we already know ‘that’
before we ask ‘why’; explanations are answers to ‘why questions’. Aristotle
(Physics, Chapter 3) wrote, ‘Men do not think they know a thing unless they
have grasped the ‘why’ of it.’ This highlights the importance of explanation.
Aristotle observes that we are often not satisfied merely knowing that some-
thing is true. We want an explanation of why it is true.
There is a long-standing literature on scientific explanation, going back
at least 60 years (see, for example, Salmon (1990: 2006), covering the first
40 years or, for a more concise treatment, Woodward (2009)). The main
competing models relate explanation to scientific laws (either universal or
statistical), to causality, and to unification.
To summarize, on the first model, a scientific explanation of an event is
a derivation of a description of the event from a scientific law (or laws),
together with some initial conditions. In this sense, to explain an event is to
show that, in some sense, the event was necessary—that it had to happen,
given the laws of nature. On a second model, to explain an event is to say
something relevant about its causal history. Lipton (1991: 2004) favors this
model of explanation, with causality broadly construed, but he is open to other
options. The third model ties scientific explanation to unification. On this
picture, proper explanations serve to unify a number of apparently disparate
phenomena, showing how they derive from a common source.
When it comes to explanation in mathematics, there are two sorts of issues.
One is the phenomenon of explanation within mathematics. The intuitive dis-
tinction between knowing that a certain mathematical proposition is true and
understanding why it is true seems to operate within mathematics as well as
in science. Some proofs are thought to be explanatory and others not so. And
Aristotle’s observation holds there as well as in natural science and in ordinary
life.
The other issue concerns cases where a mathematical fact is cited as an
explanation of a non-mathematical event. Lipton gives two examples of this,
both statistical. As he emphasizes elsewhere, there is always pressure to clarify
exactly what the explanandum is or, as he puts it, what the contrast class is.
Are we trying to explain why a given pattern tends to occur: why thrown
sticks tend to be more horizontal than vertical; why kids tend to behave in
certain ways after reward and punishment, etc. If so, then his point is that
the explanation of the pattern often involves a mathematical theorem, such as
regression to the mean, a measure-theoretic fact about spheres, or, in other
cases, the central limit theorem, a staple of contemporary statistical theory. In
each case, the theorem shows that the pattern in question is quite likely, given
some background assumptions about probabilities.
The situation is a bit more complicated if the explanandum is a single event:
why did more sticks end up horizontal this time? In principle, at least, one can
explain this in terms of the initial velocities of the sticks, the air resistance at the
time, etc., without invoking the geometrical fact that Lipton cites or anything
about probabilities. But it is surely more pragmatic, and more illuminating, to
note that the event at hand is an instance of a pattern that, according to the
underlying assumptions about probabilities and some mathematics, was very
likely to occur.
In any case, there are other, non-statistical examples where a mathematical
fact is cited in an explanation of a non-mathematical event. Here is a rather
simple-minded one: suppose that a child is given a number of rectangular tiles,
all congruent to each other, and told to make a rectangular grid with them,
using all of the tiles. After several attempts, she fails, and begins to wonder
why she cannot complete her task. The explanation is that the number of tiles
she was given is prime. For a more serious example, we all know that rain
forms into drops. Why? The explanation, it seems, involves a detour through
M AT H E M AT I C A L U N D E R S TA N D I N G : A D D E N D U M 57
the notion of surface tension, together with the mathematical fact that a sphere
is the largest volume that can be enclosed within a fixed surface area.
When it comes to mathematics, the first two models would seem to be out
of place. We do speak of mathematical ‘laws’, such as the commutative law and
the law of cosines, but that is probably only a manner of speaking. In science
and ordinary discourse, it is common to distinguish a law (such as the law of
gravity—all objects attract) from a merely accidental generalization (such as
the fact that all US Presidents before the publication of this volume were male).
That distinction does not seem operative in mathematics since, as Lipton notes,
every truth in mathematics is necessary. More importantly, the use of ‘laws’
in mathematics does not seem to play a role in distinguishing explanatory
proofs from non-explanatory ones. Presumably, everything in mathematics
is derived from ‘laws’. Nor does the notion of mathematical ‘law’ seem to
play much of a role in the citation of mathematical facts in explanations of
physical phenomena. Similarly, the notion of causality is also out of place in
mathematics. There is no sense of a mathematical proposition being the ‘cause’
of a mathematical or non-mathematical proposition.
So among the standard models for scientific explanation, only unification
seems to be applicable to mathematics. One of the main advocates of that
model, Philip Kitcher (e.g., 1989) makes that point explicitly. Mark Steiner
(1978, 1980) has written extensively on both sorts of mathematical expla-
nation, and has developed an account that is in the vague neighbourhood of
unification (but very different from Kitcher’s). That was discussed extensively
at the Symposium.
As noted, in his (2004) book, Lipton favours the causal model of explana-
tion. At the Symposium, he tentatively suggested that causality may be just one
of a family of dependence relations, and that the notion of explanation might
be tied to dependence relations generally. This proposal has the potential to
bring mathematics into the fold. The idea is that mathematical propositions
stand in some sort of objective dependency relations to each other. If this is so,
then one might maintain that explaining a mathematical proposition consists
of showing what propositions it depends on. An explanatory proof would be
one that reveals dependency relations among the premises and conclusion of
the proof, while a non-explanatory proof would show that its conclusion is true
without going via the dependency relations for the proposition.
We might go even further, and postulate that some physical events depend,
in part, on mathematical propositions, such as the aforementioned examples
concerning patterns in nature and theorems of statistics, tiling possibilities
and theorems about prime numbers, and raindrops and theorems of geometry.
If those are genuine dependency relations, then we can bring mathematical
explanations of physical phenomena into the fold as well.
Gideon Rosen’s contribution at the Symposium also invoked objective
dependency relations between propositions, including, especially, proposi-
tions in different fields, including mathematics, although Rosen’s interests are
more in metaphysics than in the understanding of explanation. There was an
extensive exchange between Rosen and Lipton, and the rest of us, on the nature
of the various dependency relations, their objectivity, and their relevance for
certain philosophical purposes.
Rosen reminded us that the eminent logician and philosopher of language,
Gottlob Frege, also invoked objective dependency relations in mathematics
(e.g., Frege, 1884: 1960). Frege often used epistemic terms for this, speaking
of the ‘justification’, ‘the proof’, or sometimes ‘the ground’ of a proposition,
but his rhetoric seems to be in the neighbourhood of present concerns. He took
the dependency relation to be objective:
. . . we are concerned here not with the way in which [the laws of number] are
discovered but with the kind of ground on which their proof rests; or in Leib-
niz’s words, “the question here is not one of the history of our discoveries,
which is different in different men, but of the connection and natural order of
truths, which is always the same”
(Frege (1884: 1960, §17); Leibniz, Nouveaux Essais, IV, §9)
Frege’s dependency relation is also asymmetric: no proposition grounds itself,

and if one proposition is a ground for another, the second is not a ground for the
first. He tried to show that arithmetic and analysis are part of logic, by arguing
that the basic propositions of those disciplines depend on general logical laws
and definitions.
We can put Frege’s project in the mould of Lipton’s and Rosen’s pro-
posal, by suggesting that Frege was looking for the proper explanation of the
propositions of arithmetic and analysis. He was not content with knowing
that the propositions are true. Scepticism and fictionalism aside (see Mary
Leng’s chapter), there is no question that we do know these propositions. Frege
was attempting to reveal why they are true. And that question is answered by
revealing the propositions that arithmetic and analytic propositions depend on.
His proposal is that the derivations of the basic principles of arithmetic and
analysis that he provides are explanatory proofs.
As indicated by the title of his book (2004), one special focus of Lipton’s
thinking is on inference to the best explanation. The idea is that a scientist will
often infer that a given proposition is true, or likely to be true, just because
it would make for the best, or as Lipton puts it, the ‘loveliest’, explanation of
some phenomenon. He closes his chapter with the suggestion that mathematics
does not deploy inference to the best explanation:
Does the idea of inference to the best explanation gain any purchase in the
context of mathematical reasoning? This might seem unlikely, since infer-
ence to the best explanation is meant to provide a partial account of non-
demonstrative reasoning, whereas at least the paradigm of mathematical
reasoning is deductive. Who needs to appeal to the wishy-washy notion of
inference to the best explanation when you have a proof? (p. 54)
He then briefly suggests that we shift from thinking about the context of
justification, and focus on the context of discovery.
M AT H E M AT I C A L U N D E R S TA N D I N G : A D D E N D U M 59
On the broadly Fregean picture sketched just above, however, we may not
have to leave the realm of justification in order to find a role for inference to the
best explanation in mathematics. The Lipton–Rosen–Frege idea is that some
mathematical propositions rest on, or depend on, others. As Frege was aware,
the regress cannot go on forever. Some propositions lie at the base of the foun-
dational enterprise, and are not explained by anything. These are the axioms.
Earlier in the essay, Lipton suggests that when it comes to proper axioms, there
are no legitimate ‘why-questions’, or at least no legitimate answers to ‘why-
questions’: ‘Thus one might hold that mathematical understanding flows from
the special epistemic status of mathematical axioms, and that these axioms are
where the regress stops.’
But how are the axioms known? The traditional view is that the axioms
are ‘self-evident’. A full and complete understanding of an axiom immediately
gives rise to justification for it. It seems to me, however, that this, traditional
foundationalist view is not really tenable when it comes to modern mathemat-
ics. Some propositions presented as axioms are hardly obvious, and it is a bit of
a stretch to say that a full and complete understanding of them will justify them
(see Shapiro, 2009). Perhaps we can say, instead, that at least some axioms
are chosen, not because of any intrinsic or self-evidence they may have, but
because they make for a good, or, as Lipton might put it, lovely explanation of
some of the theorems. It is a holistic picture. In a well-systematized branch of
mathematics, the theorems are shown to depend on the axioms. In Aristotelian
terms, the ‘why’ of the theorems lies ultimately in the axioms. When we turn
to the axioms, and ask why they are true, or at least how it is that we know
them, the answer is that they provide the best explanation of the theorems.
6
Creation and discovery
in mathematics
Mary Leng
One important role for the philosophy of mathematics is to account for the
phenomenology of the discipline, that is, to account for what it feels like to do
mathematics. One aspect of this phenomenology is the sense mathematicians
often have that they are discovering, rather than creating or inventing, the
nature of mathematical reality. Given this aspect of mathematical practice, a
natural assumption is that mathematicians are involved in the investigation
of a mathematical reality that is independent of creative human decisions,
and independent of our beliefs about that reality, in much the same way that
physical scientists are involved in the investigation of a physical reality whose
nature does not depend on us. If we accept this assumption, and the analogy
on which it is based, then the question arises: ‘What is the nature of this
mathematical reality, and how is it possible for us to have knowledge of it?’
Taking seriously the analogy with physical science would suggest that
mathematicians investigate a realm of mathematical objects, inquiring into the
nature of numbers, sets, etc., just as physical scientists inquire into the nature
of atoms. But if there is a mathematical realm of independently existing, non-
physical objects, over and above the realm of physical objects we ourselves
inhabit, then the question of how mathematical knowledge is possible becomes
pressing. Our knowledge of the physical realm stems from our interaction,
as physical beings, with that realm, but can mathematical knowledge be
accounted for in an analogous way? Mathematician G. H. Hardy described
mathematical discovery as observation of mathematical reality:
I believe that mathematical reality lies outside us, that our function is to
discover or observe it, and that the theorems which we prove, and which
we describe grandiloquently as our “creations,” are simply the notes of our
observations.
(Hardy, 1940)
But is this account of mathematical discovery as rooted in observation of a
mathematical realm tenable?
Certainly, Plato thought that knowledge of the mathematical realm could
be accounted for as resulting from a kind of direct ‘observation’. According
to Plato, mathematical objects belong to an eternal realm of Forms, which
are directly perceived by the immortal, immaterial soul prior to its ‘birth’ as
a flesh-and-blood human. In our physical incarnations, what mathematical
knowledge we have is had by recollection of this direct experience of
the Forms (Plato, Meno, 81d–86c). Our theorems then report what our
mathematical inquiry has enabled us to remember about our earlier direct
observation of the mathematical realm. However, many would find this
account of our knowledge of mathematical reality hard to swallow, requiring
as it does some powerful and potentially problematic assumptions about mind
and body. Perhaps something like Plato’s picture could be defended on the
basis of an inference to the best explanation: implausible as it may sound,
it must ultimately be accepted as the only good way of accounting for the
phenomenology of mathematical discovery. However, before opting for this
solution, it is worth examining the phenomena in question, to consider what
alternative explanations might be available.
Just what is it that mathematicians seem to discover? The discoveries
Hardy mentioned were of theorems, presumably within established math-
ematical theories. But mathematicians also, of course, create/discover new
mathematical theories, within which theorems can be proved. We must, then,
consider whether either of these kinds of discoveries are best viewed as dis-
coveries of the nature of a realm of mind-independent, non-spatiotemporal
mathematical objects.
One might think that the discovery of entirely new theories provides the
best evidence for such a mathematical realm. After all, once the axioms or
basic presuppositions of a theory are in place, our discovery of mathematical
theorems is discovery of what follows from those presuppositions, and at least
at first glance, this kind of ‘what if’ inquiry, into what would have to be true
if our mathematical axioms were true, does not require that our mathematical
axioms are in fact true of some underlying mathematical reality (though more
on this later). On the other hand, in developing new mathematical theories,
mathematicians often have a sense of discovering the basic assumptions of
these theories, as assumptions that truly describe an important corner of mathe-
matical reality, rather than simply plucking theoretical hypotheses out of the air
with a view to inquiring blithely into their consequences. Surely this aspect of
the phenomenology of mathematical practice provides the strongest evidence
for an independently existing realm of mathematical objects?
In fact, I will argue, the phenomenology of theory development is actu-
ally easier to account for from an anti-Platonist perspective than is the phe-
nomenology of mathematical proof within theories. If accounting for the
phenomenology of mathematical discovery requires us to posit any kind of
‘reality’ to ground our mathematical judgements, this reality is not a realm
of mathematical objects, but rather, I claim, a realm of objective facts about
C R E AT I O N A N D D I S C O V E R Y I N M AT H E M AT I C S 63
logical consequence. Insofar as we are concerned with understanding the sense
of discovery that is present in mathematical theorizing, the real puzzle to be
accounted for (a puzzle that, in fact, already arises even when one considers
ordinary empirical reasoning) concerns what Wittgenstein called ‘the hardness
of the logical must’ (Wittgenstein, 1953: 2001, I, §437), and not the existence
of a realm of mathematical objects.1
Let us start, though, with theory development. It is certainly true that the
choice of assumptions for a new mathematical theory is usually far from arbi-
trary, and indeed that the development of appropriate theoretical assumptions
is often rightly viewed as a significant and extremely nontrivial achievement.
But does this require us to view the development of new mathematical the-
ories as the description of an independently existing realm of mathematical
objects?
The evidence of mathematical practice, I think, speaks against this, sug-
gesting as it does constraints on our theory development that account for
our sense of discovery without requiring us to posit a realm of mathematical
objects to be discovered. For, very often, mathematical theories are developed
as solutions to problems we have set ourselves, where the constraints of the
problems are enough to narrow down the range of options that could count
as an appropriate solution (often even pinning down a unique solution). Take,
for example, W. R. Hamilton’s discovery of the quaternions which, he tells us,
‘started into life, or light, full grown, on the 16th of October, 1843’ (quoted
in Tait, 1866, p. 57). Was Hamilton’s moment of inspiration, the discovery of
the equation i2 = j2 = k2 = −1, which he excitedly carved into the stone of
Brougham Bridge, a sudden recollection of a truth contemplated by his pre-
embodied soul?
In fact, as Hamilton’s own description of his 15-year struggle to develop
rules of addition and multiplication for a three-dimensional analogue of two-
dimensional complex numbers shows, his moment of inspiration is better
viewed as a sudden realization of what had to hold, given the constraints he
had set himself. Hamilton’s aim was to discover laws for multiplying triplets
of the form x + iy + jz, along the lines of the laws for multiplying pairs x + iy,
where j was to be viewed as a square root of −1 distinct from i. A constraint he
set himself was to satisfy the ‘law of the moduli’: the modulus of the product of
two triplets should equal the product of the moduli of the two triplets taken sep-
arately. That is, if (a + ib + jc) · (x + iy + jz) = u + iv + jw, then the law of
the moduli would require that (a2 + b2 + c2 ) · (x2 + y2 + z2 ) = u2 + v2 + w2 .
This constraint, it turns out, is impossible to satisfy if multiplication is assumed
1 This is not to say that there are no aspects of mathematical practice that might
require us to assume the existence of mathematical objects. Indeed, if we turn to the
question of the applicability of mathematics, considerations of scientific confirmation
might require us to hold, not only that there are objective facts about the conse-
quences of our hypotheses concerning mathematical objects, but also that some of those
hypotheses are in fact true.
to have its usual properties of commutativity and associativity. Dropping the
commutativity constraint, however, showed Hamilton that there was room for
a solution in some special cases, so long as the product ij was equal to −ji,
and that neither were equal to 0. This effectively forced Hamilton’s hand,
leading him to discover that the only possible value for the product ij was a
third imaginary, k.
The exact problem Hamilton originally set himself had no solution. How-
ever, the system of quaternions emerged as the best way to fit as many of
Hamilton’s original constraints as possible. In fact, if we set the problem in
advance as that of extending multiplication to a system of numbers based on n
square roots of −1, preserving associativity and the law of the moduli, it is a
matter of logical consequence, rather than a mathematical ‘matter of fact’, that
only three solutions are possible, these being n = 0 (the real numbers); n = 1
(the complex numbers), and n = 3 (the quaternions).
In more pedestrian cases of mathematical theory development than Hamil-
ton’s, we often find that axiomatizations are governed by the constraint to
come up with a system of assumptions that will pin down essential features of
an already familiar mathematical or empirical system. For example, Euclid’s
axioms for geometry were intended to pin down truths about points and straight
lines in physical space, and were ‘discovered’ through examination of the ques-
tion of what had to be assumed in order to prove many other results believed
to be true of points, lines and geometric shapes. And aside from physical
interpretations, the development of axioms for mathematically familiar objects
is also commonplace—as, for example, in the development of the Dedekind-
Peano axioms, where the axiomatization is constrained by the requirement that
the structure axiomatized be an ω-sequence. Indeed, axiomatizations of mathe-
matically familiar objects often appear initially as theorems—for example, the
axioms of my own favourite corner of ‘mathematical reality’, C∗ -Algebras,
first appeared as a part of the Gelfand-Naimark theorem, which showed that
those axioms pinned down up to isomorphism subalgebras of the algebra B(H)
of bounded operators on a Hilbert space, structures which were already of
independent mathematical interest.
We are constrained, in all of these cases, by what is known already about
the system to be captured, and this constraint of course leads to the sense
that, in coming up with a formal theory, we are ‘getting something right’.
In such cases, our axioms might strike us as being ‘true’, rather than merely
convenient or interesting, but this sense of correctness is explicable even if we
do not invoke the existence of a realm of objects for our axioms to be true
of. What counts as getting something right in these cases is dependent on the
constraints we have set ourselves (which require us to pin down a structure
satisfying various assumptions). We need not (though we may) think of these
constraints as being imposed by an independent realm of objects about which
our theories must assert truths. Rather, as consequences of a collection of initial
assumptions or constraints, new axiomatizations can come about in much the
way that theorems do within the context of already established theories. In each
case, the area of mathematical discovery that really matters seems to be the
discovery of consequences of one’s assumptions.
Let us turn, then, to the kind of mathematical theorizing that takes place
against a backdrop of accepted assumptions (such as axioms). In the paradigm
cases, such reasoning is deductive, and amounts to the proving of theorems
from axioms, although there is also some room for the use of abductive reason-
ing in such contexts: mathematicians may reason that, given their assumptions,
such and such a result is likely to be true. Sticking for now with the central
case of deductive proof from axioms, we can consider what conclusions can
be drawn from the sense mathematicians often have that in such activity they
are involved in discovery rather than creation of mathematical results. Are
mathematicians who are engaged in proving theorems really discovering the
already determined consequences of their assumptions, or could it be the
case that, despite the strong sense of discovery, they are actually involved in
the creation of links between axioms and theorems that were not already, in
some sense, ‘out there’? If mathematicians are engaged in discovery rather
than creation, what implications does this have for our view of the nature
of mathematics. In particular, is this discovery discovery of a realm of mind-
independent mathematical objects? On the other hand, if we do choose to view
deductive proof from axioms as a matter of creation, rather than discovery, can
this be reconciled with the felt objectivity of mathematical proof, and, indeed,
the applicability of mathematical reasoning?
A natural line of thought takes it that, yes, deductive mathematical rea-
soning is objective, leading us to the discovery of logical consequences of
our mathematical assumptions. But such objectivity has nothing to do with
an independent realm of mathematical objects, but is entirely a result of the
objectivity of logic. After all, in reasoning to a theorem P on the basis of math-
ematical assumptions A1 , . . . An , we prove not that P is true, but rather, that
if A1 & . . . & An , then P. This conditional claim does not assert the existence
of any mathematical objects; its truth, we may suppose, rests solely on the
fact that P is a logical consequence of A1 , . . . , An . According to this way of
thinking, there is nothing particularly problematic about the felt objectivity of
mathematical proof from axioms—it is simply a special case of the objec-
tivity of any deductive reasoning, and depends solely on the objectivity of
the ‘following from’ relation. Furthermore, as we have seen, a similar story
can be told about the kind of reasoning that leads to the development of new
mathematical theories: although this reasoning does not start from axioms, it is
nevertheless governed by logical constraints established by the consequences
of our preformal mathematical assumptions and/or desiderata.
There is, however, a problem with this comfortable-seeming position,
which arises once we consider what we mean by the claim that P follows
logically from A1 , . . . , An . We surely do not mean by this that P can be
derived from A1 , . . . , An using an accepted collection of rules of inference. For
one thing, we know of cases where this analysis falls short of capturing our
usual notion of logical consequence: take the second-order Peano axioms for
arithmetic. Gödel’s first incompleteness theorem tells us that, for any (stan-
dard) collection of inference rules we can come up with, there will be a
sentence G in the language of (2nd order) Peano arithmetic which follows
logically from those axioms but which is not derivable using our chosen
collection of rules. But even without the worries that Gödel’s theorem brings,
one should be wary of an analysis of logical consequence which rests facts
about logical consequence on a chosen collection of inference rules; after all,
what makes a collection of inference rules a good one is, presumably, that those
rules respect facts about logical consequence, and not vice versa.
All of which is simply to say that the relevant notion of logical consequence
underlying the objectivity of mathematical reasoning is that of semantic, rather
than syntactic consequence. In the relevant, semantic, sense, P is a logical
consequence of A1 , . . . , An if and only if it is not logically possible for P to
be false while A1 , . . . , An are all true. But this analysis simply replaces one
undefined logical notion (logical consequence) with another (logical possibil-
ity). If our question is what grounds the objectivity of logical consequence,
then surely an analogous question arises for logical possibility: can we say
anything more about what the logical possibility or impossibility of a sentence
amounts to?
Here is where the difficulty for our comfortable view of the objectivity
of mathematical reasoning arises. For, arguably, the best analysis of logical
possibility available is mathematical: a sentence P is logically possible if there
is a set theoretic model in which that sentence is interpreted as a truth. On this
analysis, P is a logical consequence of A1 , . . . , An if and only if, in all models
which make A1 , . . . , An true, P is also true. If this analysis is correct, then
the existence of objective facts concerning logical consequence comes down
to the existence of a realm of mathematical objects (set theoretic models).
So the objectivity of mathematical discovery is, after all, dependent on an
objective realm of mathematical objects, and we are led back to the difficulty
of explaining our knowledge of such things.2
Is there an alternative analysis of logical possibility available? One might
try to eschew abstract mathematical objects in favour of logically possible
concrete worlds. But even if we could make sense of the notion of a logically
possible world in such a way as to respect our intuitions concerning logical
possibility, if it is facts about these worlds that ground the objectivity of math-
ematical inference, we will still have difficulty explaining how we could have
knowledge of the following-from relation, since such worlds are presumably
spatiotemporally isolated from our own. Since we do seem to know some
facts about what follows from what, we should be wary of any analysis of the
2 Georg Kreisel reportedly maintained that the question of realism in mathematics

amounts to ‘the question of the objectivity of mathematics and not the question of
the existence of mathematical objects.’ (Putnam, 1975: 1979, p. 70). If this analysis
of mathematical objectivity is correct, then these two questions cannot, after all, be
separated.
following-from relation that grounds such facts in matters which seem forever
beyond our grasp.
Such concerns might lead us to abandon all attempts to reduce logical
possibility to something more basic. Indeed, drawing from a discussion of
Kreisel’s (1967), Hartry Field (1984: 1989, 1991), has argued that we should
view logical possibility as a distinct notion from the related formal notions
of deductive and model theoretic consistency. We can learn about logical
possibility via derivations and models: we know that, if (in an accepted deriva-
tion system) we can derive a contradiction from a sentence S, then S is not
logically possible, and that, if (in an accepted set theory) we can find a model
in which the sentence S is interpreted as a truth, then S is logically possible.
But (on the Kreisel/Field view), logical possibility is a distinct notion from
the related deductive and model theoretic notions, and should not be thought
of as reducible to either. Rather, Field suggests, we should see ‘it is logically
possible that’ as a unary logical operator that is no more in need of ‘reduction’
than is the unary operator ‘it is not the case that’. Both, Field thinks, should
be explicated through specification of their inferential role, rather than via a
reduction to something more primitive. Note, then, that this account accepts
the existence of irreducible modal facts grounding the objectivity of mathe-
matical reasoning. While avoiding commitment to the existence of abstract
mathematical objects, this account still requires us to accept a kind of reality
underpinning our mathematical discoveries (albeit a ‘realm’ of modal facts,
rather than of abstract mathematical objects). And once more we will need to
ask what it is about us as humans that allows us to have knowledge of these
modal facts.
But perhaps there is another response to the phenomenology of math-
ematical discovery: perhaps we can accept the feeling that our judgements
concerning logical consequence have an objective ground that is independent
of human decisions, but hold that this sense of objectivity is nevertheless an
illusion. This is the approach that Wittgenstein takes in his conventionalist
approach to mathematics. According to Wittgenstein, despite appearances,
‘The mathematician is an inventor, not a discoverer’ (Wittgenstein, 1956:
1978 I, p. 167). In proving mathematical theorems, we do not discover the
consequences of our mathematical hypotheses, but rather, decide to accept the
conclusion proved as a new consequence in our theory. Far from teasing out
the content of our mathematical concepts,
the proof changes the grammar of our language, changes our concepts.
It makes new connections, and it creates the concept of those connections.
(It does not establish that they are there; they do not exist until it makes them.)
(Wittgenstein, 1956: 1978, III, p. 31)
Perhaps, then, despite appearances, there are no objective facts about logical
consequence in mathematics, just the results of human decisions that could
always have gone differently?
Perhaps this view could be sustainable if individual mathematical theories
were entirely isolated from one another, so that ‘decisions’ made within one
theory would not clash with decisions made elsewhere. Indeed, Wittgenstein
himself thinks that cross-theoretical links are themselves a matter of decision,
that, for example, it is a matter of choice to embed the natural numbers in the
integers and so on (see, e.g., Waismann, 1979, pp. 34–6). But here the phenom-
enology of mathematical discovery speaks strongly against the conventionalist
position, as Friedrich Waismann notes in explaining (in his (1982) paper,
‘Discovering, Creating, Inventing’) his own abandonment of Wittgensteinian
conventionalism. There are just too many examples of theorems proved in
one mathematical context bearing out (and even illuminating) the conclusions
drawn in other areas. Waismann’s example is of a result concerning the real
numbers that receives an explanation once the reals are embedded in the
complex numbers. The Taylor series expansion
1
= 1 − x2 + x4 − x6 + · · ·
1 + x2
converges for |x| < 1, but diverges for all other real values of x. Once we
1
embed the reals in the complex numbers, the behaviour of the function 1+z 2 in
its real portion is explained by the fact that the complex function has singular-
ities at z = ±i, together with a theorem of complex analysis which tells us that
any power series expansion only converges within a circle of radius R about the
origin, and diverges elsewhere. Given these facts about complex functions, the
real-valued function could not have behaved other than it did. Far from being
a matter of human convention, such results seem determined independently of
our choices. It is, as Waismann remarked, as if the real function already knew
that the complex numbers were there.
Related to this issue (which we might call the applicability of mathematics
within mathematics), is the phenomenon of the applicability of mathematics
to nonmathematical questions. We are abundantly aware of cases where math-
ematical reasoning is used to derive empirical predictions, and where these
predictions turn out to be correct. One view on the applicability of mathematics
takes the applicability of mathematics to reside in structural similarities: a
mathematical theory is (sometimes) applicable to a nonmathematical phe-
nomenon because the nonmathematical reality is structurally similar to some
portion of the structure described by the mathematical theory in question. But
if in mathematical reasoning we were simply freely deciding, at each step,
what we will take to be true of the objects of a given mathematical theory,
it is surely entirely mysterious how these free decisions so regularly result in
accurate predictions.
Both of these phenomena, then, speak against the radical anti-objectivist
account of mathematical reasoning. What, then, does the phenomenon of
mathematical discovery have to tell us about mathematical reality? Not, I think,
that our mathematical theorems are answerable to an independent realm of
mathematical objects. But if we do not accept Wittgenstein’s extreme conven-
tionalism, at the very least we must accept that our mathematical discoveries
are underpinned by objective facts about logical consequence. And if we wish
to hold, with Kreisel, that the problem that concerns us is ultimately ‘not
the existence of mathematical objects, but the objectivity of mathematical
statements’ (Dummett, 1978, p. xxviii), then we will have to accept that the
relevant facts concerning logical consequence do not reduce to facts about
mathematical models. If we want to understand mathematical discovery, then,
we must consider from where the objectivity of these facts might arise.
Comment on Mary Leng’s ‘Creation
and discovery in mathematics’
Sensing objectivity
Michael Detlefsen
It is not uncommon for those experienced in doing mathematics to see it as an

activity of discovery or observation rather than innovation or creation. G. H.
Hardy and Kurt Gödel were among prominent twentieth-century mathemati-
cians who believed so.
Leng agrees with Hardy and Gödel that a sense of discovery is a signif-
icant part of our mathematical experience. She believes in addition, though,
that the most convincing ‘felt objectivity’ is one concerning facts of logical
consequence.
Things might rest there but for the fact that, in Dr. Leng’s view, the most
compelling treatment of logical consequence is that given by contemporary
model theory, and judgements concerning models involve us as much in diffi-
culties concerning knowledge of abstract objects as views of ‘felt objectivity’
not restricted to logical consequence. She thus concludes that the ‘felt objectiv-
ity’ of mathematics cannot unproblematically be attributed to the objectivity of
logical consequence. This notwithstanding, she does not see a clearly prefer-
able alternative.
This may under-represent the difficulties of accepting the ‘felt objectivity’
of logical consequence as a datum for mathematical epistemology. In the first
place, it does not address the possibility of different conceptions of logical
consequence. If we include practising intuitionists and other kinds of con-
structivists among those having relevant mathematical experience, there will
not be agreement on what the compelling instances of logical consequence are.
Indeed for some (e.g., Brouwer), there will not be agreement on the importance
of any judgements of logical consequence, intuitionist or classical, to what is
rightly counted as mathematical thinking.
Greater attention might also have been given to what it is that is supposedly
felt to be objective in judgements of consequence. Leng says it is semantical
C R E AT I O N A N D D I S C O V E R Y I N M AT H E M AT I C S : C O M M E N T 71
fact, but there are other possibilities too. As a mathematical activity, proving
is intended to bring about a certain response in a certain audience. That this is
so suggests that judgements of consequence ought ultimately to reflect what
the prover judges to be the inferential standards of the intended audience.
Whether those standards are best specified in semantical terms or in terms
of comportment with trusted non-semantically specified rules is a matter on
which there is room for deep disagreement.
Finally, the complexities of ‘felt objectivity’ may be underestimated. Gödel
wrote of propositions ‘forcing’ themselves on us as being true. Is this what
‘felt objectivity’ is supposed to come to? Or is ‘forcedness’ only a part of ‘felt
objectivity’? If the former, then felt objectivity does not seem to provide much
evidence for mind-independence. Innate dispositions and even conditioning
brought about by training can surely give rise to feelings of being ‘forced’. In
his Grundgesetze (Frege, 1903: 1962, II, §142), Frege rightly warned against
taking such feelings of compulsion as indications of truth. As he noted, one
‘need only use a word or symbol often enough, and the impression will be
produced that this proper name stands for something; and this impression will
grow so strong in the course of time that in the end hardly anybody will have
any doubt about the matter.’
If felt objectivity is only forcedness, then, it is not a potent indicator of truth
or objectivity. If, on the other hand, it is more, then we need to be told what
its other elements are and how they give surer evidence of mind-independence
than mere forcedness does.
7
Discovery, invention and realism:
Gödel and others on the reality
of concepts
Michael Detlefsen
Introduction
This chapter is an investigation into the question whether there are features
of our acquisition of mathematical knowledge that support a realist attitude
towards mathematics. More particularly, it is a reflection on the reasoning
which moves from the claim that
I. mathematicians are commonly convinced that their reasoning is part of
a process of discovery, and not mere invention,
to the claim that
II. mathematical entities exist in a noetic realm to which the human mind
has access.
For convenience, I’ll refer to this as the original argument.
The use of the term ‘noetic’ in II calls for brief comment. Traditionally it
has been used to signify a type of apprehension, noēsis, which is characterized
by its distinctly ‘intellectual’ nature. This has generally been contrasted to
forms of aisthēsis, which is broadly sensuous or ‘experiential’ cognition, or
intuition. There is interest, and difficulty, in determining more exactly the ways
in which an intellectual experience of a supposedly non-sensuous reality might
resemble and might also contrast with sensuous experience of material objects.
This is where a great deal of the difficulty concerning the content of ‘noetic’
will be met. The terms of such a comparison will therefore be one of our chief
concerns.
Experience and involuntariness: background
For Plato, the objects of noēsis were the Forms, which he believed to be
manifested by experience while also transcending it. The empiricists, on the
other hand, generally associated intellectual apprehension with apprehension
of concepts or ideas. These, when legitimate, were mental representations
obtained by abstraction from sensory experience.
Kant famously emphasized a distinction between two types of representa-
tions, intuitions (Anschauungen) and concepts (Begriffe). Among their impor-
tant differences, he maintained, was one concerning the extent to which they
are within the power of the judging agent to control. Concepts were taken to
be spontaneous (see Kant, 1781: 1990, A50–51/B74–75), or capable of being
brought into existence by a judging agent’s own intellectual initiative (selbst
ausgedachten, op. cit., A639/B667). Intuitions were not. In the end, however,
Kant required that genuine or legitimate concepts be consistent (nicht selbst
widersprechen, op. cit., A150/B189). The freedom to create or generate them
was thus a constrained freedom.
This notwithstanding, Kants saw a great difference between our control
over concepts and our control over intuitions (see op. cit., A19/B33, B132).
He regarded intuitions and their relations as given (gegeben) (see op. cit.,
A19/B33, B132) and not under the spontaneous productive control of our
minds. Concepts, on the other hand, could be thus produced. There was there-
fore no guarantee that they be exhibited by any object(s).
. . . even if our judgment contains no contradiction, it may connect concepts in
a manner not borne out by any object, or in a manner for which no ground is
given . . . and so may still, in spite of being free from all inner contradictions,
be either false or groundless.
Kant (1781: 1990, A150; B190)1
Kant’s acceptance of this asymmetry between concepts and intuitions provides
an interesting point of contrast to more recent views of the nature of concepts.
Specifically, it seems to be sharply at odds with the view of concepts presented
by Gödel in various foundational writings of the 40s, 50s and 60s (see Gödel,
1947: 1990). It is to Gödel’s view(s) that I now turn.
Gödel agreed with Kant in regarding the involuntariness of a representation
as a mark of its objectivity. However, whereas Kant regarded our use of
mathematical concepts as essentially creative or voluntary,2 Gödel regarded
it as distinctly involuntary. Contrary to Kant, he thus maintained that
. . . despite their remoteness from sense experience we do have something like
a perception also of the objects of set theory, as is seen from the fact that the
1 See Kant (1781: 1990, Bxxvi) for similar remarks.

2 As Kant put it:
. . . I can think (denken) whatever I want, provided only that I do not con-
tradict myself, that is, provided my concept (Begriff) is a possible thought
D I S C O V E R Y, INVENTION AND REALISM 75
axioms force themselves upon us as being true. I don’t see any reason why we
should have less confidence in this kind of perception, i.e., in mathematical
intuition, than in sense perception.
Gödel (1947: 1990, p. 268)
Indeed Gödel claimed more than the involuntariness of our apprehension
of concepts. He believed that it has a perception-like, but still non-sensory
character. He maintained as well that the cognitions it yields are in important
ways independent of both the voluntary acts and involuntary dispositions of
our minds.
I am under the impression that . . . the Platonistic view is the only one tenable.
Thereby I mean the view that mathematics describes a non-sensual reality,
which exists independently both of the acts and the dispositions of the human
mind and is only perceived, and probably perceived very incompletely, by the
human mind.
Gödel (1951: 1995, pp. 322–23)
Despite these realist convictions, however, Gödel conceded certain points to
the conventionalist. Specifically, he believed, they were right to think that
mathematics is about concepts rather than physical or psychical items (see
Gödel, 1951: 1995, p. 320). They were right too, he said, to believe that
mathematical truths in some sense owe their truth to the meanings of terms—
specifically, to the concepts expressed by terms (loc. cit.). Where the con-
ventionalist went wrong, he believed, was in taking these meanings to be
determined by conventions (ibid.). The truth as he saw it was rather that
these concepts form an objective reality of their own, which we cannot create
or change, but only perceive and describe.
Gödel (1951: 1995, p. 320)
In Gödel’s view, then, mathematical concepts are discovered and not cre-
ated by acts of convention or other mental acts or dispositions. Similarly for
truths concerning them (loc. cit.). The chief evidence of this, in his view,
was the involuntariness with which truths concerning mathematical concepts
‘force’ themselves on us as being true. This, to Gödel, was signal indication
of their independence from our creative capacity, a capacity he took to mainly
(möglicher Gedanke). This suffices for the possibility of the concept (Begriff),
even though I may not be able to answer for there being, in the sum of all
possibilities, an object (Objekt) corresponding to it. Indeed, something more
is required before I can ascribe to such a concept objective validity (objektive
Gültigkeit), that is, real possibility (reale Möglichkeit); the former possibility
is merely logical. This something more need not be sought in the theoretical
sources of knowledge (theoretischen Erkenntnisquellen); it may lie in those
that are practical.
Kant (1781: 1990, Bxxvi, note *)
be constituted by our mental dispositions and our various abilities to perform
voluntary mental acts.
Gödel’s phenomenological argument

Gödel’s variant of the original argument thus emphasized what he took to be a
broad phenomenological feature of our mathematical experience—namely, the
involuntariness of our basic mathematical knowledge. Unlike the version of the
original argument set out at the beginning, it attached little if any significance
to the mere fact (if it is a fact) that mathematicians are commonly convinced
that they discover rather than invent.
The core element of his argument was thus the claim that
1. Propositional contents concerning mathematical concepts are imposed
or ‘forced’ on us as being true in a manner similar to that in which
propositional contents are impressed on us by sensory experience.
He seems further to have believed that
2. This imposed character is best explained by seeing it as a consequence
of a perception-like experience of a realm of beings whose existence and
characteristics are independent of our mental acts (e.g., acts of convention
or stipulation) and dispositions, both individual and generic.3,4
From this, he suggested, we may rightly infer that
3. Mathematical concepts exert a non-sensory cognitive influence on us,
and their existence and properties are independent of our mental acts and
dispositions.
The most plausible account of our overall mathematical experience thus
implies that
4. Mathematical beliefs are about objectively existing things that we
discover rather than invent or create (e.g., by acts of stipulation or
convention).
This, in sum, is Gödel’s argument, an argument which provides one type
of more extended articulation of the original argument. What I find most
interesting and distinctive about it is its appeal to a supposed phenomenon
3 Gödel expressly denied, however, that our perception-like experience of mathe-

matical truth was or was at bottom based on sensory perception. It was perception-like
only in that it was given to or forced upon us in a manner akin to that in which the
contents of sensory perception are given to or forced upon us.
4 In saying that mathematical concepts (at least some of them) are ‘independent’
of our mental acts and dispositions, Gödel seems to have meant that (i) mathematical
concepts exist and would continue to exist even if our mental acts and dispositions did
not, and (ii) no change in our mental acts or dispositions would automatically result in
a change in the properties of those concepts.
of ‘forcedness’ concerning our mathematical judgements (or at least some
of them), and the idea that this phenomenon serves as evidence for the
external reality of mathematical concepts. In the next section I’ll consider
Gödel’s reasoning more carefully and also attempt to clarify certain ways
in which it differs from earlier arguments for the reality of mathematical
entities.
‘Forcedness’ as an indicator of reality

Gödel’s language, especially his statement that mathematical propositions
‘force themselves upon us as being true’ (Gödel, 1947: 1990, p. 268) suggests a
broadly phenomenological type of reasoning. More specifically, it is reasoning
that takes its lead from what it supposes to be our experience of the evidentness
of certain propositions.
Gödel seems to have been particularly concerned with the evidentness of
basic or primitive mathematical propositions—propositions the forcedness of
which does not seem to be explicable by their being perceivedly implied by
(other) forced propositions. This at any rate is what he seems to have had
in mind when he noted not only that certain propositions of set theory force
themselves on us as being true, but that its axioms do.5
Gödel seems to have thought that proper sensitivity to the particulars of
our experience of evidentness in mathematics will reveal that at least some
propositions are ‘forced on’ us in a manner similar to that in which sensory
propositions are ‘forced on’ us by sensory experience. He seems to have seen
this, moreover, as indicative of an external mathematical reality to which we
have broadly ‘experiential’ access.
There exists, unless I am mistaken, an entire world consisting of the totality of

mathematical truths, which is accessible to us only through our intelligence,
just as there exists the world of physical realities; each one is independent of
us, both of them divinely created.
Gödel (1951: 1995, p. 323)
5 Gödel’s exact statement was ‘the axioms [of set theory] force themselves upon us
as being true.’ ‘The’ axioms? Was Gödel assuming that there is a unique, or perhaps
uniquely best, axiomatization of set theory? Not necessarily. Given a set-theoretic
language L, there is no necessary incoherence in believing that (i) there is an identifiable
set of propositions formulable in L whose evidentness does not derive from that of
other propositions formulable in L, and also that (ii) not all elements of need or
even ought to be taken as axioms of an axiomatization of set theory. There might, for
example, be logical overlap between elements of that would make it unnecessary or
even undesirable to take all of them as axioms of an axiomatization of set theory. There
might also be alternative ways of thinking about sets, or different concepts of set, that
would divide the elements of in such ways as would associate certain elements of
with certain concepts or ways of thinking and not others.
Such reasoning, of course, raises many questions. Among these are:
Q1: How reliable and revealing an indicator of external reality is our expe-
rience of the forcedness of sensory propositions?
Q2: How reliable and extensive is the asserted analogy between the forced-
ness of basic mathematical propositions and the forcedness of basic
sensory propositions as indicators of external realities?6
Q1 queries the evidential connection between the property of sensory
judgements called forcedness and the existence of an external source respon-
sible for it. Affirmation of such a connection would seem to require belief that
forcedness of sensory judgement is best explained by some sort of transfer
of energy—specifically, transfer to a sensory agent from a sense-stimulative
source external to her.
Understood this way, is phenomenal forcedness of sensory judgements a
reliable indicator of external reality? Here, I think, we need to distinguish
two types of such reliability. One is what I’ll call existential reliability, or
reliability concerning the existence of an external source for an experience
of forcedness. The other is what I’ll call attributive reliability, or reliability
concerning the characteristics of a supposed external source forced on us in a
sensory judgement.
Exceptions to both existential and attributive reliability are, of course,
familiar from the literature on sense perception and I’ll not go into them in
any detail here. Rather, I’ll simply note that quasi-sensory experiences such as
certain hallucinations provide cases where an experience of forcedness is not
an existentially reliable indicator of an external reality. Similarly, well-known
cases of optical illusion raise similar concerns regarding attributive reliability.
In addition to these concerns, there are three others I’ll mention. The first
has to do with our understanding of ‘forcedness’. What did Gödel mean when
he said that our mathematical experience includes the experience of proposi-
tions’ forcing themselves upon us as being true? A natural interpretation would
include the following implication: P’s being forced upon us as true implies that
we form a belief that P.
This raises an important question concerning Gödel’s supposed mathemat-
ical ‘perception’. The reason why is that sense perception does not seem to
sponsor the above implication. That is, having a sensory experience as of P
(e.g., an experience as of one line segment’s being longer than another) does
not seem to imply that we form a belief that P. There are well-known illusions
(e.g., the Ponzo and Müller-Lyer illusions) in which we experience one line
segment’s being longer than another but do not believe it to be so.
6 More basic than either Q1 or Q2, of course, is the difficult question of how to make
sense of the notion of a sensory proposition. I won’t address this problem here, though,
since it is Q1 and Q2 and their focus on the phenomenon of forcedness that are my
chief concerns.
In sensory perception, then, we can have an experience as of P but not
judge or believe that P. Sensory perception seems to be a form of apprehension
in which not all contents it presents are presented as true. Is the same true of
Gödel’s mathematical ‘perception’? That is, can mathematical perception in
some sense ‘give’ appearances which, despite there being given in this way,
are nonetheless not forced on us as being true?
I don’t know the answer to this question, but it’s not the answer that is
my chief concern here. Rather, it is the question itself. It shows, if I am not
mistaken, a certain type of difficulty posed by Q2, a difficulty that will confront
any attempt such as Gödel’s to exploit an analogy between sense perception
and a perception-like form of apprehension in mathematics.
Nor is this the only such difficulty or the most serious. More troublesome,
I think, are problems raised by cases of sensory illusion where false contents
are forced on us as being true. A well-known illustration is Adelbert Ames’
‘Distorted Room’. Viewed in such a way as to eliminate stereoptic information
(e.g., viewed through a peephole), the room looks like an ordinary ‘cubical’
room with rectangular windows, a flat, rectangular, level floor and rectangular
walls of equal and uniform height and depth.
When commonplace objects (e.g., an adult human being of normal size and
a smaller child of normal size) are placed on opposite sides of the room and
photographed, and the photographs are merged into a single image, strange
appearances occur. The child, for example, will appear to be much larger than
the adult.
The truth, of course, is that Ames’ room is not, despite its appearance, an
ordinary cubical room. Its seemingly rectangular windows, walls and floor are
trapezoidal, its walls are not of uniform height and depth and the floor is not
level. That it looks normal, and continues to look normal when familiar types
of objects are placed in it suggests the influence that dispositions can have on
perception.7 As R. L. Gregory described it:
Evidently we are so used to rectangular rooms that we accept it as axiomatic
that it is the objects [human bodies] which are odd sizes rather than that the
room is an odd shape. But this is essentially a betting situation—it could be
either, or both, which are peculiar. Here the brain makes the wrong bet, for the
experimenter has rigged the odds. Indeed . . . the most interesting feature of
the Ames Distorted Room is its implication that perception is a matter of mak-
ing the best bet on the available evidence . . . wives do not see their husbands
as distorted by the Room—they see their husbands as normal, and the room
its true queer shape . . . [Also] Familiarity with the room, especially through
touching its walls . . . does gradually reduce its distorting effect on other
objects, and finally it comes to look more or less as it is—distorted in fact.
Gregory (1969, pp. 180–181, brackets added)
7 Gödel, it will be recalled, maintained that ‘mathematics describes a non-sensual

reality, which exists independently both of the acts and the dispositions of the human
mind and is only perceived’ (Gödel, 1951: 1995, p. 323, emphasis mine).
Sense perception can therefore in some sense and in some circumstances
‘force’ false contents on us as being true. It does so, moreover, through the
operation of deep-seated mental dispositions and not just the objective proper-
ties of the material objects involved.
Any attempt to establish an analogy between sense perception and a
perception-like form of mathematical apprehension must thus address the
question whether there are parallels to such phenomena as Ames’ ‘Distorted
Room’ in the case of mathematical perception. Specifically, it must consider
whether (i) false as well as true contents can be forced upon us as being
true, whether (ii) the phenomenon of forcedness is any surer indication of
the existence and characteristics of external objects than it is of the existence
and working of deep-seated mental dispositions, whether (iii) there is a phe-
nomenon of ‘familiarity’ or ‘training’ that affects mathematical perception
in a way similar to that in which sensory familiarity/training affects sensory
judgement in the case of the ‘Distorted Room’,8 whether (iv) there is a sys-
tematic connection between reliability, on the one hand, and belief-selection
that favours such familiarity/training, on the other, and, finally, whether (v)
there are reliable phenomenal means of distinguishing forcedness which arises
from the influence of external objects from forcedness which arises from the
operation of mental dispositions.
In addition, there are other problems as well. One is to discern when an atti-
tude that registers as acceptance is indeed a case of accepting a propositional
content as true. There are different types, degrees and modes of acceptance,
and different objects of acceptance. Mere awareness of what we identify as an
attitude of acceptance would therefore seem to offer little guidance concerning
its exact character and the ultimate source of its compulsion. Much of what
we experience as compelled acceptance is doubtless due to long, complex and
largely invisible conditioning of various sorts, and we are generally far from
8 The possibility of an effect of familiarity/training on belief-selection in mathemat-

ics should, I think, be taken seriously. I lack the space to explore this idea further here,
but it seems to resonate with various widely advocated constraints on mathematical
theorizing. Foremost among these is the so-called Principle of Permanence, a principle
formulated and energetically defended by Peacock and others in the early and mid-
nineteenth century, and a principle widely adopted by later mathematicians. A relatively
recent statement endorsing (one variant of) the Principle of Permanence is the following
by Courant and Robbins:
. . . the essential logical and philosophical basis for operating in an extended

number domain is formalistic; that extensions have to be created by defini-
tions which, as such, are free, but which are useless if not made in such a way
that the prevailing rules and properties of the original domain are preserved
in the larger domain.
Courant and Robbins (1947: 1981, p. 89)
For more on the Principle of Permanence see Detlefsen (2005, pp. 273–278).
knowing its nature and origins. Frege had a particularly desultory view of the
conditioning that he took to lie behind many of our affirmations.9
One need only use a word or symbol often enough, and the impression will
be produced that this proper name stands for something; and this impression
will grow so strong in the course of time that in the end hardly anybody has
any doubt about the matter.
Frege (1903: 1962, Vol. II, §142)
Neither Gödel nor others since his time have adequately addressed these
difficult problems. And though this should not be taken to suggest that they
are irresolvable, it should motivate consideration of alternative approaches to
Gödel’s seeming understanding of the invention/discovery question in mathe-
matics. It is to the consideration of such alternatives that I now turn.
Traditional alternatives
Gödel’s view shares a concern with the more important traditional views con-
cerning the discovery vs. invention issue. Broadly speaking, this is a concern
for the objectivity of mathematics. A little more exactly, it is a concern to
restrict certain types of subjectivity in mathematics, specifically those having
to do with the possibility of legitimately using ‘created’ or ‘invented’ concepts.
I’ll now briefly survey some of the more interesting parts of the historical
literature that touch on this issue.
Ancient views
The extent to which mathematics should be open to creation or invention
was a common concern in antiquity. Looking at the ancient literature, though,
we quickly notice a terminological difference. Specifically, what is currently
meant by ‘discovery’ is quite different from what the ancients meant.
This can be seen from Cicero’s (106 BC–43 BC) use of the term. He
distinguished two types of methods proper to careful, systematic inquiry in
any area of thought, mathematics included. Methods of the one type he called
methods of ‘discovering’ (Cicero, 1894–1903, p. 459), methods of the other
type, methods of ‘deciding’ (ibid.). In time this grew into the traditional
division between artis inveniendi (arts of invention, arts of discovery or arts of
investigation) and artis iudicandi (arts of adjudication, also sometimes called
artis demonstrandi (arts of demonstration)).
This distinction was important in science and also in jurisprudence.
In mathematics, it generally took the form of a distinction between
9 In the concluding section I’ll discuss two others, Bolzano and Dedekind, who
also emphasized the elaborate conditioning that underlies even ordinary mathematical
beliefs.
discovermental and demonstrative methods of investigation. The end of the
former was the efficient development of new knowledge, albeit only perhaps
of sub-optimal quality. The end of the latter was the perfection of knowledge—
specifically, the perfection of the imperfect knowledge produced by the justi-
ficatively sub-optimal methods of discovery.10
Traditionally, then, ‘discovery’ and ‘invention’ were synonyms. They
marked the first stage of a classical two-staged conception of justification
which was divided into a certificative stage and an argumentative stage. The
general idea behind this Classical Scheme can be summarized as follows:
Genuine scientific justification of a proposition11 requires both certifica-
tion and argumentation. Argumentation by itself is not enough because
it can too easily be fictional. De-fictionalization of argumentation is
the proper job of certification. Genuine demonstration, then, is certi-
fied argumentation, and genuine scientific knowledge requires genuine
demonstration.
This scheme figured significantly in the thinking of ancient geometers. An
interesting example is Proclus (411–485) who, in his Commentary on Book
I of the Elements, appealed to the Classical Scheme to explain the ‘ordering’
of Propositions I–IV.12
10 It was commonly believed as well that pre-demonstrative investigation ought to

be preparative to demonstrative investigation in the sense of making the development
of demonstrations easier or more efficient. Archimedes (287 BC–212 BC), for example,
recommended ‘mechanical’ methods to Eratosthenes as means of furthering the search
for demonstrations.
. . . I have thought fit to write out for you and explain in detail. . . a certain
method, with which furnished you will be able to make a beginning in the
investigation by mechanics of some of the problems in mathematics. . . this
method is no less useful even for the proof of the theorems themselves. For
some things first became clear to me by mechanics, though they had later to
be proved geometrically owing to the fact that investigation by this method
does not amount to actual proof; but it is, of course, easier to provide the
proof when some knowledge of the things sought has been acquired by this
method rather than to seek it with no prior knowledge.
Archimedes (1993, pp. 221–222)
11 That is, justification capable of sustaining genuine scientific knowledge (scientia,

epistēmē).
12 Propositions I–IV are:
I: On a given finite straight line to construct an equilateral triangle.

II: To place at a given point (as an extremity) a straight line equal to a given straight
line.
III: Given two unequal straight lines, to cut off from the greater a straight line equal
to the less.
. . . our geometer [Euclid, Bk. 1] follows up these problems [Props. I–III]
with his first theorem [Prop. IV] . . . For unless he had previously shown the
existence of triangles and their mode of construction, how could he discourse
about their essential properties? Suppose someone. . . should say: ‘If two
triangles have this attribute, they will necessarily also have that.’ Would it not
be easy. . . to meet this assertion with “Do we know whether a triangle can be
constructed at all?”. . . It is to forestall such objections that the author of the
Elements has given us the construction of triangles. . . These propositions are
rightly preliminary to the theorem. . .
Proclus (1970, pp. 182–183)
Like other ancient geometers, Proclus’ conception of the distinction

between certificative and argumentative methods followed the lines of the
ancient distinction between problematic investigations and theorematic investi-
gations. The former consisted in ‘the working out of problems’ (Proclus, 1970,
p. 157), the latter in ‘the discovery of theorems’ (loc. cit.).
The aim of the former was ‘to produce, bring into view, or construct what in
a sense does not exist’ (loc. cit.), and accomplishment of this aim was generally
secured by ‘construct[ing] a figure, or set[ting] it at a place, or apply[ing] it to
another’ etc. (loc. cit.). In sum, it was generally accomplished by construction.
The aim of theorematic investigation, on the other hand, was ‘to see, iden-
tify, and demonstrate the existence or nonexistence of an attribute’ (loc. cit.). In
other words, it was an effort ‘to grasp firmly and bind fast by demonstration the
attributes and inherent properties belonging to the objects that are the subject-
matter of geometry’ (op. cit., pp. 157–158).
Problematic investigations were thus intended to establish the existence
of something ‘new’ (i.e., something whose existence had not previously been
established), while theorematic investigations were intended to establish the
properties of already existing things. Hence the need for prior certification of
the existence of the subjects of theorematic investigations.
Proclus’ comment on the ordering of Propositions I–IV reflects acceptance
of such a scheme. Proposition IV is a theorem whose subject is triangles, lines,
equal lines, etc. Propositions I—III, on the other hand, are problems, which
were needed in order to establish the existence of the subject of Proposition IV.
They were thus ‘rightly preliminary’ (op. cit., p. 183) to it even though they
were not used in the proof for it.13
IV: Two triangles having two of their respective sides equal, and the angles con-
tained by those sides equal, will also have equal bases, be equal to each other as
triangles, and have the remaining angles equal.
13 Proposition IV is also about equal triangles, angles and equal angles. This being
so, one would naturally expect to see preliminary propositions establishing the exis-
tence of these items. Proclus did not comment on this and we can therefore only wonder
whether he regarded it as a defect in Euclid.
In saying that Propositions I–III are ‘rightly preliminary’ to Proposition
IV, Proclus seems to have meant that they address certain broad sceptical
challenges to it and/or its proof. Proposition I, for example, responds to the
challenge raised by the question ‘Do we know whether a triangle can be
constructed at all?’. Propositions II and III, on the other hand, respond to the
challenge that perhaps no straight line (or other geometrical quantity for that
matter) is equal to another.14
Overall, then, Proclus seems to have taken the view that fabricated theo-
rematic investigation is uncertifiable investigation or investigation that lacks a
subject. To avoid it, the objects that are to form the subject-matter of a would-
be theorem should be priorly shown to exist. This, Proclus believed, was (best?
only?) done by giving a genetic or constructive definition of the subject—a
definition which ‘explains the genesis of a thing; that is, how the thing is made
or done: as is this definition of a circle, viz., that it is a figure described by the
motion of a right line about a fixed point’ (Hutton, 1795–1796: 2000, Vol. 1,
p. 362).15
Modern views
Mathematicians and philosophers of the modern era held similar views, as the
following statement by Leibniz illustrates:
. . . the concept of circle put forward by Euclid—namely, that it is the figure
described by the motion of a straight line in a plane about one fixed end—
affords a real definition, for it is clear that such a figure is possible. It is useful
. . . to have [such] definitions. . . beforehand. . . [For] we cannot safely devise
demonstrations (secure texere demonstrationes) about any concept, unless we
know that it is possible; for of what is impossible or involves a contradiction
(impossibilibus seu contradictionem involventibus), contradictories can also
be demonstrated. This is the a priori reason why real definition is required for
possibility.
Leibniz (1683:1973, pp. 12–13 (294), brackets added)16
14 More specifically, Propositions II and III provided two basic methods for produc-
ing equal lines: one when no line previously exists at a given place, the other when a
longer line exists at that place (see Proclus, 1970, p. 183).
15 Similar ideas are familiar from western jurisprudence, in particular, from its
general adoption of the principle of corpus delicti. This requires evidence not only
for justified conviction but for justified trial as well. The idea is that for a trial to be
justified there must be evidence both of the ‘existence’ of a crime and of the identity
of its perpetrator. In the case of murder, this has typically taken the form of evidence
of death as the result of an act, and evidence of the criminal agency of an identified
person as its means. Other types of crimes call for other types of evidence of course.
All legitimate trial, however, requires evidence of both the existence of a crime and the
identity of the criminal.
16 The number in parentheses is the page number of the Latin text in Leibniz (1978,
Vol. 7). For a similar statement see Leibniz (1989, XXIV).
Leibniz thus held that genetic or constructive definitions are valuable
because they make known a priori the possibility of the concepts they define
(see Leibniz, 1764: 1916, Bk. II, ch. II, §18). This in turn, he suggested, makes
demonstration ‘safe’ in a certain way.
He made a confusing variety of claims concerning impossible concepts and
their existence, however. Sometimes he suggested that there are impossible
concepts (i.e., concepts that imply a contradiction).
A concept is either suitable or unsuitable. A suitable concept is one that is
established to be possible, or not to imply a contradiction.
Leibniz (1903, p. 513)
Other times he suggested that impossible concepts cannot exist. Thus, for
example, he claimed that ‘we cannot have any idea of the impossible’ (Leibniz,
1978, Vol. 4, p. 424), and he also maintained that ‘what actually exists cannot
fail to be possible’ (Leibniz, 1978, Vol. 7, p. 214).17
My main concern here, though, is with the suggestion that there is some-
thing unsafe about demonstration concerning A when it is not accompanied by
knowledge of the possibility of A. Specifically, I’m concerned with the reasons
Leibniz may have had for believing this.
The danger he mentioned was a possibility of contradiction—‘of what is
impossible . . . contradictories can be demonstrated’ (loc. cit.). Taken as we
would take it today, though, this does not seem right. The principal type of
theorem in Leibniz’ time was a proposition of the form ‘All A are B’. The
contradictory of this would be (a sentence equivalent to) a sentence of the form
‘Some A are not B’, which requires the existence of A’s.
When A is impossible, though, A’s cannot exist. Hence, one who, like
Leibniz, maintained that the actual must be possible, would not have believed
it possible to demonstrate ‘Some A are not B’ for impossible A. Consequently,
he would have had no reason to regard demonstration of ‘Some A are not B’
as posing a threat to demonstration of ‘All A are B’ when A is impossible.
More likely, Leibniz was thinking of demonstration of ‘No A are B’
rather than demonstration of ‘Some A are not B’ as the imagined counter-
demonstration to demonstration of ‘All A are B’. Supposing this to be right,
though, we are left with the problem of saying what it is about such a situation
that would make it ‘unsafe’. Whatever else Leibniz may have thought the
danger to be, it would not be one of directly demonstrating a false statement.
That this is so follows from elementary (classical) logical facts. When A is
impossible, there are necessarily no A’s, and when this is so ‘All A are B’ and
‘No A are B’ are both true.18
17 See Mates (1986, pp. 66–68) for a brief but useful discussion of these puzzling
matters.
18 This supposes of course that A may still be a genuine concept even when it is
impossible. There were, of course, many who challenged this, on which more later.
But if it is not the falsehood against which real definition of A protects
us, from what does it protect us? Leibniz was not explicit on this point.19 He
may have seen the threat as a threat of fictionalization—that is, the threat of
engaging in investigation that has no subject and hence, in the final analysis,
is about nothing and produces knowledge of nothing. This seems to have been
the classical view and was essentially the view Proclus was urging. Or so I
would argue. Others of Leibniz’ time, and of earlier and later times too, were
more definite, and they generally gave one of two views of the security issue,
one broadly practical, the other broadly theoretical in character. I’ll presently
describe and discuss each of these. For the moment, though, the main points to
bear in mind are that (i) there was and perhaps still is a tradition in the history
of mathematics of linking the reliable with the broadly experiential, at least to
the extent that construction was taken to be a medium of experience, but that
(ii) the content of such experience was not necessarily that of the proposition
ultimately being proved or justified.
Real definition as a practical concern

The practical defense of real definition did not challenge the common idea that
the reality or possibility of a concept is constituted by its consistency. Its claim
was rather that the only practical means of establishing the consistency of a
concept was to exhibit an instance of it, that is, to give a real definition of it.20
The efficacy of real definitions as a means of ensuring consistency was
commonly seen as the reason why classical geometry, in which real definitions
predominated, was so much less problematic than algebra, in which they did
not. The following remark by Playfair is a typical eighteenth-century expres-
sion of this thinking:
The propositions of geometry have never given rise to controversy, nor needed
the support of metaphysical discussions. In algebra, on the other hand, the
doctrine of negative quantity and its consequences have often perplexed the
19 Leibniz’ example of a real definition was Euclid’s definition of a circle—namely,

that it is ‘the figure described by the motion of a straight line in a plane about one fixed
end’ (loc. cit.). This was in keeping with the general understanding of real definition
in the seventeenth, eighteenth and nineteenth centuries. Hutton, for example, gave the
following characterization in his Dictionary:
Real Definition . . . explain[s] the genesis of a thing; that is, how the thing is
made or done: as is this definition of a circle, viz., that it is a figure described
by the motion of a right line about a fixed point.
Hutton (1795–1796: 2000, Vol. 1)
20 By the ‘consistency’ of a concept is meant the consistency of the statements that

are taken to describe it or to apply to it.
analyst, and involved him in the most intricate disputations. The cause of
this diversity must no doubt be sought for in the different modes which they
employ to express our ideas. In geometry every magnitude is represented by
one of the same kind; lines are represented by lines, and angles by an angle,
the genus is always signified by the individual, and a general idea by one of
the particulars which fall under it. By this means all contradiction is avoided,
and the geometry is never permitted to reason about the relations of things
which do not exist, or cannot be exhibited.
Playfair (1778, pp. 318–319)
The practical identification of demonstration of consistency with produc-

tion of an instance became, if anything, even more definite in the nineteenth
century. The following statement by John Herschel expressed the common
attitude:
The test of truth by its application to particulars being laid aside, noth-
ing remains but its self-consistency to guide us in its recognition. But this
in axiomatic propositions amounts to no test at all . . . Their [the axioms’]
mutual compatibility as fundamental elements of the same body of truth, can
only be shown by experience—by the observed fact of their coexistence as
literal truths in a particular case produced.
Herschel (1841, p. 220, brackets added)
Frege too pressed the point in his disagreement with the creativists (e.g.,
Hankel and Hilbert). He thus responded to Hankel’s identification of consis-
tency and existence by observing;
Strictly, of course, we can only establish that a concept is free from contradic-
tion by first producing something that falls under it. The converse inference
is a fallacy, and one into which Hankel falls.
Frege (1884:1968, §95)
Similarly, in his correspondence with Hilbert concerning his (Hilbert’s)

seeming reliance on the possibility of proving the consistency of a concept
without providing a verifying instance of it, Frege wrote:
What means have we of demonstrating that certain properties, requirements
(or whatever else one wants to call them) do not contradict one another? The
only means I know is this: to point to an object that has all those properties,
to give a case where all those requirements are satisfied. It does not seem
possible to demonstrate the lack of contradiction in any other way.
Frege to Hilbert, January 6, 1900 (Gabriel et al., eds., 1980, p. 43)
Eventually, frustrated by Hilbert’s intransigency, he challenged him to

explain how it is possible to prove the consistency of a concept other than
by identifying a verifiable instance of it.
I believe I can deduce, from some places in your lectures, that my arguments
failed to convince you, which makes me all the more anxious to find out
your counter-arguments. It seems to me that you believe yourself to be in
possession of a principle for proving lack of contradiction which is essentially
different from the one I formulated in my last letter and which, if I remember
right, is the only one you apply in your Foundations of Geometry. If you were
right in this, it could be of immense importance, though I do not believe in it
as yet, but suspect that such a principle can be reduced to the one I formulated
and that it cannot therefore have a wider scope than mine. It would help to
clear up matters if in your reply to my last letter—and I am still hoping for a
reply—you could formulate such a principle precisely and perhaps elucidate
its application by an example.
Frege to Hilbert, September 16, 1900 (Gabriel et al., eds., 1980, p. 49)
Frege’s reference to Hilbert’s Grundlagen der Geometrie was to his proof

of the consistency of his system of geometry by interpretation in the real
numbers (see Hilbert, 1899, §9). He was right to note that this was the only
proof of consistency Hilbert offered at the time. It would be another twenty
years before his proof-theoretic conception of consistency was even outlined.21
That production of a verifying instance was the only way to establish
the consistency of concept was thus a prevalent belief of the seventeenth,
eighteenth and nineteenth centuries. It is also one which allows us to explain
why Leibniz and others should have seen demonstration of ‘No A are B’ as
posing a threat to demonstration of ‘All A are B’. The explanation is as follows.
If both ‘All A are B’ and ‘No A are B’ are provable, then, supposing that what
is provable is true, there will be no A’s. If, however, there are no A’s, there will
in practice be no way of establishing the consistency of all that is eventually
asserted of A (including, ex hypothesi, ‘All A are B’). The only practical way
of doing this is to produce an instance of A and to verify of that instance that it
has the properties attributed to it by the assertions that have it as their subject-
concept. In the end, then, we need instantiation of A because we eventually
will need to prove the consistency of the statements made about A, and, as a
matter of practical necessity, this can only be done by producing an A.
The supposed upshot of this reasoning for the question whether mathe-
matics is created or discovered is this: thoroughgoing creativism regarding
concepts is unsustainable. The reason is that every theory must be consistent,
and, in the final analysis, this can only be shown by identifying a witness-
ing instance. Such witnesses, Arnauld observed, can only be discovered, not
created.
21 In my view it is unclear how closely a model-construction proof like Hilbert’s

resembles the type of constructive or concrete instantiation that Leibniz and Frege
seemed to take real definition to provide. The model Hilbert offered for geometry was
not one which persuades through visualization or experience. It’s far more abstract
than Euclid’s real definition of a circle and it presumes an ability to make abstract
judgements. That model-construction of the type Hilbert used is really by intuition, or
is more discursive in character, and, so itself in need of intuitive vindication seems open
to doubt.
. . . nominal definitions are arbitrary but real definitions are not. Since any
sound is . . . capable of expressing any idea whatever, I am permitted for
my own use to choose a certain sound to express . . . one idea. But the case
is completely different with real definitions, for the relations which obtain
among ideas are independent of man’s will.
Arnaold (1964, p. 83)
The question of the proper roles of nominal and real definition were thus
at the heart of the traditional discussion of the discovery vs. creation issue. It
remained a central element well into the twentieth century. Understanding this,
I believe, should help us both to deepen and to broaden our own perspective on
the issue. More particularly, it should lead us to an appreciation of what more
may lie behind the issue, and what more may be at stake in its resolution than
arguments like Gödel’s suggest.
Real definition as a theoretical concern

The theoretical defence of real definition is based on a view of how we acquire
concepts. This is a broadly empiricist or experientialist view which sees
abstraction from experience as the fundamental means by which concepts are
developed. It therefore classifies concepts as real or not according to whether
there is a plausible account of how they might have arisen, and whether an
abstractive path from experience is also specified. Concepts which arise in this
way are legitimate or real, those which do not are not.
This view has had many advocates among mathematicians and philoso-
phers from the modern era onward. John Leslie’s popular early nineteenth
century geometry text, for example, taught that ‘Geometry is. . . founded on
external observation, but such observation is so familiar and obvious that
the primary notions which it furnishes might seem intuitive, and have been
regarded as innate’ (Leslie, 1809, p. 2). And again ‘Geometry, like the other
sciences which are not concerned about the operations of the mind, rests
ultimately on external observation’ (op. cit., p. 453).
A common synonym for ‘real’ among broadly empiricist thinkers of the
eighteenth and nineteenth centuries was the term ‘given’, a term Leslie char-
acterized as follows in the sequel to the text cited above:
Quantities are said to be given, which are either exhibited, or may be found.
Leslie (1821, p. 4)
Overall then, the theoretical defence of real definition in geometry was based
on two ideas. The first was that geometrical concepts are derived from obser-
vation of external objects by a process(es) of abstraction. The second was that
the contents of such observation are in an important sense given rather than
created. Real concepts are thus representations that begin with a content that
is given, not one that is created or fabricated. They then arise by a process of
abstraction from this experience, a process which only subtracts from and does
not add to the original experienced contents.
This view of real concepts was sometimes extended to a general view
of concepts by nineteenth-century philosophers. A striking example was
Schopenhauer, who offered the following memorable account of the nature
of concepts and the conditions under which they exist.
. . . concepts derive their content from the intuitive realm and therefore the
entire structure of the world of thought rests upon intuitions. We must there-
fore be able to go back from every concept, even if indirectly through inter-
mediate concepts, to the intuitions from which it is itself abstracted . . . That
is to say, we must be able to support it with intuitions which stand to the
abstractions in the relation of examples. . .
These intuitions. . . afford the real content (realen Gehalt) of all our thought,
and whenever they are wanting we have not had concepts but mere words in
our heads (blosse Worte im Kopfe). In this respect our intellect is like a bank
which holds notes (Zettelbank), which, if it is to be sound, must have cash in
its safe, so as to be able to meet all the notes it has issued, in case of demand.
The intuitions are the cash and the concepts the notes.
Schopenhaver (1859: 1966, Book 1, ch. 7, vol. 2)22
Views of this general type were not, however, confined to philosophers or to
the nineteenth century. Indeed, no lesser nor less recent a mathematician than
Hermann Weyl used the image of notes of deposit to describe the difference
between genuine concepts and propositions, on the one hand, and purely
symbolic devices, on the other. He used the existence claim in mathematics
as his chief example, and he likened it to paper money—that is, money which
in itself has no value and whose only true value is that of a real commodity that
backs it.
An existential proposition (Existentialsatz)—something like ‘there is an even
number’—is not at all a judgment in the genuine sense of an assertion of
a fact. Existential states of affairs (Existential-Sachverhalte) are an empty
invention (Erfindung) of the logician. ‘2 is an even number’: that is a real
(wirkliches), a factual expression of a given judgment. ‘There is an even
number’ is only a judgment-abstract (Urteilsabstrakt) obtained from this
judgment. Just as I take knowledge (Erkenntnis) to be a valuable (wertvollen)
holding (Schatz), so I regard the judgment-abstract as paper (Papier), which
somehow indicates (anzeigt) a holding without disclosing (verraten) where
it is. Its only value can lie in its ability to get me to search for the hold-
ing. The paper is worthless so long as it is not realized (realisiert) by
a real (wirkliches) judgment such as ‘2 is an even number’ that stands
behind it.
Weyl (1921, p. 54)
22 See Schopenhauer (1911, p. 76) for German version. The first edition was pub-
lished in 1819, the second in 1844, a third in 1859. The passage quoted here is in all
these editions.
In all the above we see the same basic idea at work, namely, that those
representations we call concepts exist only to the extent that they stand in a
relationship of abstractability to the contents of experiences that are given and
not created.
This offers a relatively clear explanation of how it is that proof of ‘No A are
B’ would trivialize or devalue proof of ‘All A are B’. If both ‘All A are B’ and
‘No A are B’ expressed propositional contents and were true, there would be
no A’s. Hence there would be no experiential content—that is, no experienced
instance of A—from which A could be abstracted. The expression ‘A’ would
therefore have no content and would not express a concept. As a result, ‘All
A are B’ would not express a proposition and could not therefore be a content
of genuine judgement. As a result, a demonstration of ‘All A are B’ would
not yield knowledge that all A are B, and, so, would not achieve its intended
purpose.
But though this reasoning is clear, it is also unconvincing, and for at least
two reasons. The first is its foreclosure of the possibility of uninstantiated
concepts. This is troubling since there seem to be genuine concepts (or at
least concept-like representations) that not only are not instantiated but, indeed,
are not capable of being instantiated. This at any rate for composite concepts
(e.g., that of a map requiring five colors for its colouring). Whether there are
concepts that can properly be regarded as primitive and are not instantiable is a
more difficult question to answer. I know of no good argument that there could
not be such contents, however.
The second reason why the Schopenhauer–Weyl argument is unconvincing
is its seeming blindness to the possibility that the type of experience capable
of sustaining derivation of a concept by abstraction might itself presuppose
the availability of concepts. The Schopenhauer–Weyl argument supposes that
we could have the types of contents needed to begin a process of abstraction
without having access to concepts that are not based on such acquisition. To
be convincing, though, a closer analysis of contents and abstraction would
have to be offered, and, with it, an argument that there are at least some types
of rudimentary contents we can grasp without prior application of concepts.
In addition, reasons would have to be given for the claim that there is a
plausible path of acquisition from such contents to mathematical contents via
abstraction.
Until such gaps are filled, the theoretical defence cannot readily be
accepted as a general account of the existence or acquirability of concepts.
But though it has shortcomings as a general theory of mathematical concepts,
the theoretical defence might yet provide a basis for a worthwhile distinction
between discovery and creation in mathematics. The idea would be that a real
concept is an instantiable concept. The existence of non-instantiable concepts
would not be denied. Neither would their use be forbidden. Their use, however,
would be controlled by an appropriate form of consistency requirement—
specifically, a consistency requirement the demonstrated satisfaction of which
would not necessitate instantiation of the concept(s) involved.
Discussion
Both of the traditional defences of real definition discussed above offer possi-
bilities of significant distinctions between discovery and creation in mathemat-
ics. Both also diverge from Gödel’s phenomenological argument in that they
see the distinction between discovery and creation as wanted for more than
merely a phenomenologically adequate accounting of whatever experience we
may have of truth being ‘forced’ on us. It is not necessary that they deny that
there is a point behind Gödel’s phenomenological reasoning. Neither, though,
do they see it as indicating the primary role for experience in the development
of mathematical knowledge.
Both defences take the view that producing an instance of a concept is
fundamentally a matter of discovery rather than of invention. Moreover, they
both see discovery as discovery of reality where reality guarantees consistency
or possibility. Creation, on the other hand, does not. In order for real definition
to fulfill its intended justificative purpose, then, it is necessary that it be seen as
a case of discovery rather than creation. Otherwise, the mathematician would
possibly be faced with an endless regress of proper justificative demands, and
this is surely something she should want to avoid.
I mentioned above that the practical defence of real definition seems to
underestimate the variety of forms a consistency constraint on the use of cre-
ated concepts might take. This was due to the common belief prior to Hilbert’s
development of proof theory that the only way to prove the consistency of a
concept was to find a recognizable instance. Hilbert’s proof-theory provided a
syntactical alternative to this.
On the other hand, Gödel’s incompleteness theorems (particularly his sec-
ond theorem) suggest that there are limits to the applicability of the Hilbertian
alternative. This may mean that, practically speaking, we will often have to rely
on production of instances or construction of models to prove consistency. In
the end, then, Frege may have been right concerning the practical possibilities
of proving consistency.23 Therefore, in the final analysis, it is difficult to
determine how extensive a role for discovery the practical defence is capable
of supporting.
Conclusion
I would like to close by considering one final question concerning Gödel’s
proposed use of the phenomenon of given-ness or forcedness in mathematics.
This time, though, my concern is not whether we can reliably detect it and, if
so, with what degrees of confidence and discernment. Rather, it is with what a
proper response to experiences of ‘given-ness’ or ‘forcedness’ might or ought
to be in the first place, supposing that we do in fact have them. I was led to this
23 There are of course serious questions concerning the similarity of model-

construction to the kind of exhibition of an instantiating instance supposed to typify
real definition. I lack the space to go into these here, however.
by Bolzano’s trenchant questioning of the role and character of real definition
in mathematics.24
As is well-known, Bolzano’s proof of the intermediate value theorem was
largely the product of his concern that the methods used to prove a theorem
be of a level of generality appropriate to that of the theorem proved. In his
view, a general theorem about quantity ought to appeal only to general laws
of quantity, and never to laws pertaining only to some particular type of
quantity.
This led him to reject previous proofs of the intermediate value theorem on
account of the appeals they made to specifically geometrical quantities. It also
led him to reject the usual proofs of geometrical theorems, many of which he
believed to follow from general laws of quantity. A more accurate picture of
the true grounds of such theorems would be given by proofs whose premises
were themselves general laws of quantity.
. . . in Euclidean geometry no spatial object is accepted as real unless its con-
struction has first been demonstrated by means of plane, circle and straight
line. This restriction betrays its empirical origin clearly enough. Board, com-
pass, and ruler are. . . the simplest instruments which were needed initially
for drawing. However, considered in themselves the straight line, the circle,
and also the plane are such compound objects that their possibility cannot
be accepted in any way as a postulate. . . For example, the proposition that
between every two points lies a mid-point is far simpler than the proposition
that between every two points a straight line can be drawn. Nevertheless,
Euclid proves the former from the latter and several others. It is sufficient for
the theoretical exposition of mathematics. . . that one proves the possibility of
every conceptual connection which is put forward. How, and in what way,
an object analogous to the concept can be produced in reality belongs to
practical mathematics.25
Bolzana (1810: 2004, §37)
In my view, part of what Bolzano was doing here was to point out dangers of
following the leadings of forcedness. Too much trust may lead to such things
as proving the more elementary (though perhaps less forced) from the less
elementary (but more forced).
This raises a related question concerning Gödel’s views. He said that it
is the axioms of set theory that force themselves on us as being true. At the
24 Dedekind raised similar concerns in arithmetic. In the first paragraph of Was sind
und was sollen die Zahlen? he thus warned against following the leadings of inner
intuition (innere Anschauung), and said that evident truths are quite often those most in
need of proof. This indeed became the leitmotif of this foundational work in arithmetic,
the central principle of which was that nothing at all capable of proof should be accepted
without proof, regardless of the degree of its evidentness. See Dedekind (1888), preface
to the first edition.
25 In this sentence, the reality of which Bolzano speaks is empirical or sensible
reality. He thus demotes the type of real definition found in Euclid from scientific
mathematics to what was then regarded as applied mathematics.
same time, he likened forcedness in mathematics to sensory perception. This
would lead us to think that the axioms of set theory are to the larger domain of
mathematical truth what sense perceptions are to the larger domain of natural
scientific truth.
If this is correct, though, then, pursuing the analogy between mathematics
and natural science, the propositions of set theory that force themselves on
us as being true are not likely to be the basic laws of mathematics. That is,
they should not properly be taken to be axioms. Rather, they should be seen
as phenomena whose proper explanation requires deeper and more basic laws,
a theoretical mathematics if you will.26 It is not easy, however, to think what
these deeper laws might be. In addition, the analogy between the axioms of set
theory and sensory judgements does not square well with what Gödel himself
said about the relationship of set theory to number theory—namely, that the
former can be at least partially justified by its ability to simplify proofs of
theorems of the latter.
Overall, then, there are serious problems concerning the analogy between
sensory experience and the type of experience by which Gödel supposed
various truths of mathematics to force themselves on us as true. If forcedness
is symptomatic of experience, then the mathematical propositions forced on
us as true ought to be seen as the experiential part of mathematics. Pursu-
ing Gödel’s analogy we would then be led to consider the possibility of an
observation/theory divide in mathematics, similar to that which we accept
in natural science. Bolzano believed in something like this. Specifically, he
believed in objective differences of basicness between mathematical proposi-
tions. Moreover, he took it as the principal duty of the mathematician to reveal
this ordering of relative basicness.
Bolzano thus raised important questions concerning given-ness and its
possible significance to mathematical epistemology. I’ve indicated some of
the challenges these questions raise for Gödel’s attempt to introduce a type
of experience into mathematical epistemology. These should warn us against
equating degree of given-ness or forcedness with degree of basicness. They
do not, so far as I can see, speak with similar force against the traditional
understanding of the discovery/creation issue in mathematics—namely, that
which centers on the use of so-called ‘real’ definition in mathematics and the
protection it may provide against certain of the grosser forms of subjectivity.
26 Russell and others, of course, expressly adopted this ‘regressive’ conception of

the axioms of set theory.
Comment on Michael Detlefsen’s
‘Discovery, invention and realism’
John Polkinghorne
Michael Detlefsen gives us a careful discussion of the claim made by Kurt

Gödel that we have experience of being confronted by mathematical objects in
a non-negotiable manner, similar to the confrontation with physical objects that
persuades us of the independent reality of the physical world. He is surely right
to assert that this claim, if it can be substantiated, provides the best ground for
the defence of the conviction held by many mathematicians that their reasoning
is a matter of discovery rather than the mere invention of pleasing intellectual
puzzles.
In probing this analogy, I think it is necessary to take fully into account
the subtlety of our encounter with the physical world. Defenders of realist
claims for modern physics, of whom I am one (Polkinghorne, 1996, ch. 2),
are not simply appealing to the fact that we bump into large objects. The
quantum world, for example, is something much too veiled and elusive in its
character to be treated as if it were simply naively objective. Nevertheless,
physicists are convinced of the reality of entities such as electrons or quarks,
refusing to treat them as if they were just imaginative and imaginary devices to
enable calculations to be made. I believe that the defence of physical realism
ultimately depends upon the intelligibility that these entities enable us to
attain. We take electrons and quarks with ontological seriousness because their
existence explains great swathes of more directly accessible phenomena. The
power of group theory to illuminate the structures of symmetrical patterning
(du Sautoy, 2008) seems to offer an analogy in favour of taking the noetic
reality of finite groups seriously.
Another argument in favour of the reality of the physical world is its
character of quite often proving surprising, with properties contrary to prior
‘reasonable’ expectation. Quantum theory is the prime exemplar of this. This
resistance to prior expectation is persuasive that we are encountering an inde-
pendent reality standing over against us. I suppose that the nineteenth-century
discovery of non-Euclidean geometries would be a mathematical analogy.
Detlefsen recalls Kant’s distinction between intuitions, which come to us
with the inexorable character of giveness, and inventions, which he believed
can be created and manipulated at human will. It seems to me that there is
some connection here with an experience testified to by mathematicians, of
the coming into consciousness of a deep mathematical theory ‘fully-formed’,
so to speak. In my own chapter I referred to a famous anecdote of how Henri
Poincaré, after months of struggle with Fuchsian functions, found that the
complete theory sprang into his mind as an unbidden gift just as he was
setting off on holiday. The non-negotiable character of mathematical ideas
seems illustrated by the fact that a mathematician considering a particular
axiomatized system can ‘see’ the truth of the Gödelian sentence, despite its
being formally unprovable in that system.
Einstein once said that the basis of fundamental physics has to be ‘freely
invented’. He certainly was not subscribing to a postmodern idea of the subjec-
tive creation of merely pleasing notions. Einstein was too uncompromisingly
objective in his thinking for that. Rather, he was pointing to the role of the
leap of creative imagination that enabled him, for example, to write down the
beautiful equations of general relativity, even if they then had to be verified
by comparison with phenomena. Great ideas in mathematics are similarly
grasped which display the property of being ‘deep’—that is, an apparently
simple formulation proves to have extensive and unexpected consequences,
in analogy to the range of prediction of a successful physical theory when it
is found to imply consequences not even empirically known when the theory
was discovered. Think of the astonishing fruitfulness of the idea of complex
numbers.
Defences of realist interpretations in both physics and mathematics have to
be subtle and delicate and it seems to me that the two disciplines are cousins
under the skin in this respect.
8
Mathematics and objectivity
Stewart Shapiro
I wish to explore, in a tentative and general way, the extent to which mathe-
matics is objective. As is typical in philosophy, part of our question is to try
to get clear on the meaning of the terms in the question. I take it for granted
that we know what mathematics is, or at least that we know it when we see
it, borderline cases aside. But what of ‘objectivity’? Intuitively, to be objective
is to be independent of human judgements, conventions, forms of life and the
like. But what of this notion of ‘independence’?
The view that mathematics is not objective—that mathematical truths are
somehow tied to the nature of human cognition, conventions or whatever—is
not uncommon among philosophers. Immanuel Kant took mathematics to flow
from ‘pure intuition’, the form of our faculty of perceiving the world spatially
and temporally. Thus, mathematics is directly connected to human abilities,
and so lacks objectivity—at least in some sense of that term. The traditional
intuitionists, L. E. J. Brouwer and Arend Heyting, follow suit. Heyting (1931:
1983, p. 52) wrote:
The intuitionist mathematician proposes to do mathematics as a natural func-
tion of his intellect, as a free, vital activity of thought. For him, mathematics
is a production of the human mind . . . [W]e do not attribute an existence
independent of our thought, i.e., a transcendental existence, to . . . mathemat-
ical objects . . . [M]athematical objects are by their very nature dependent on
human thought.
A prominent, contemporary philosopher, Terrence Horgan (1994, p. 99)

adopts a convention, due to Hilary Putnam, that words written in small cap-
itals are meant to be ‘about denizens of the mind-independent, discourse-
independent, world’. It would beg the present question if this supposed mind-
independent, discourse-independent world was thoroughly material, and nei-
ther Horgan nor Putnam assumes it is.
Horgan (1994, pp. 100–101) claims that
there is a whole spectrum of ways that a sentence’s correct assertibility
can depend upon THE WORLD. At one end of the spectrum are sentences
governed by . . . norms . . . [such that a sentence in those discourses] is true
only if some unique constituent of THE WORLD answers to each of its
singular terms . . . At the other end of the spectrum are sentences whose
governing assertibility norms . . . are such that those sentences are sanctioned
as correctly assertible by the norms alone, independently of how things are
with THE WORLD.
He then adds, parenthetically that ‘sentences of pure mathematics are plausible

candidates for’ this second status, of being assertible or not ‘independently of
how things are with THE WORLD’. So Horgan takes it as ‘plausible’, without
further comment, that pure mathematics lacks any objectivity.
Galileo Galilei’s 1623 book Il Saggiatore (The Assayer) contains the much
quoted passage:
Philosophy is written in this grand book—I mean the Universe—which stands
continually open to our gaze, but it cannot be understood unless one first
learns to comprehend the language and interpret the characters in which it
is written. It is written in the language of mathematics, and its characters are
triangles, circles and other geometrical figures, without which it is humanly
impossible to understand a single word of it; without these, one is wandering
in a dark labyrinth.
Today, we are in an even better position to appreciate the deep truth underlying
this than they were in the early decades of the seventeenth century. There is
hardly a branch of natural or social science that does not have substantial
mathematical prerequisites. One cannot get beyond the first few pages of a
textbook without considerable mathematical facility. Of course, this is applied
mathematics, while Horgan comments on pure mathematics. But the difference
between the more theoretical parts of some of the sciences and pure mathemat-
ics is not all that clear, at least to me.
Views, like Horgan’s, that deny the objectivity of even pure mathematics
make a mystery of Galileo’s observation—in a pejorative sense of ‘mystery’. If
mathematics is governed by no more than human conventions of assertibility,
as Horgan suggests, then why is mathematics so important for science? There
must be something about the WORLD, as it is in itself, independent of human
concerns, judgements, etc., that puts mathematics at the centre of just about all
of our attempts to understand it—even if they are our attempts to understand
the universe.
Galileo speaks of the language of the universe. Theology aside, I presume
that this is a metaphor. More literally, the claim is that one must invoke
mathematics in order to fully or properly understand the universe, at least in a
scientific manner. Does it follow that mathematical assertions themselves are
objective? It is probably also true that one cannot understand the universe at
any sophisticated depth without understanding a language. Does it follow that
M AT H E M AT I C S AND OBJECTIVITY 99
languages themselves are directly tied to the nature of the world, objectively,
independent of human interests, concerns, judgements, and the like? In any
case, there must be something about the non-linguistic world that makes
languages what they are, and make them effective in communication for us
humans. Similarly, there must be something about the non-mathematical world
that makes mathematics effective—indeed essential—for us to understand just
about anything.
To be fair, Heyting and Horgan are surely correct that mathematics and,
for that matter, language and science, are human activities, and the pursuit and
results of those activities are shaped by human concerns and interests. It is a
truism that theories and explanations, in both mathematics and science, are due
to both the nature of the non-human world and the nature of human knowers
and understanders. As John Burgess and Gideon Rosen (1997, p. 240) put it,
‘our theories of life and matter and number are to a significant degree shaped by
our character, and in particular by our history and our society and our culture.’
Of course, this is not to say that the world itself or, as Horgan or Putnam
might put it, THE WORLD ITSELF, is somehow shaped by ‘our character’. The
Burgess–Rosen observation is that it is our theories of the world that are so
shaped. The question before us is the extent to which the truths of mathematics
are due to the way the non-human world is, and the extent to which these truths
are due to the way the human mathematicians are.
Several competing philosophical traditions have it that there is no way to
sharply separate the ‘human’ and the ‘world’ contributions to our theorizing.
As Protagoras (supposedly) said, ‘man is the measure of all things’. On some
versions of idealism, not to mention some postmodern views, the world itself
has a human character. The world is shaped by our judgements, observations,
etc. And so, it would seem, there just is no WORLD, in Horgan’s and Putnam’s
sense. A less extreme position is Kant’s doctrine that the ding an sich, (or
DING AN SICH) is inaccessible to human inquiry. We approach the world
through our own categories, concepts, and intuitions. We cannot get beyond
those, to the world (or THE WORLD) as it is, independently of said categories,
concepts and intuitions.
On the contemporary scene, a widely held view, championed by W. V. O.
Quine, Hilary Putnam, Donald Davidson, and Burgess, has it that, to use a
crude phrase, there simply is no God’s eye view to be had, no perspective from
which we can compare our theories of the world to the WORLD itself, to figure
out which are the ‘human’ parts of our successful theories and which are the
WORLD parts.
This Kant–Quine orientation may suggest that there simply is no objectiv-
ity to be had, or at least no objectivity that we can detect. If this is right, then
there simply is no answering the question of this paper. For what it is worth,
I would resist this, despite sympathy with the Kant–Quine orientation. There
may not be such a thing as complete objectivity—whatever that would be—
but it still seems that there is an interesting and important notion of objectivity
to be clarified and deployed. There seems to be an important difference—a
difference in kind—between statements like ‘pure water contains two hydro-
gen atoms for each oxygen atom’ and statements like ‘Broccoli is disgusting’
and ‘Manchester United is evil’. Our question concerns which side of this
divide contains mathematics.
Crispin Wright’s Truth and Objectivity (1992) contains an account of
objectivity that is more comprehensive than any other that I know of, providing
a wealth of detailed insight into the underlying concepts. Wright does not
approach the matter through metaphysical inquiry into the fabric of REALITY,
wondering whether THE WORLD contains things like moral properties or num-
bers. He focuses instead on the nature of various discourses, and the role that
these play in our overall intellectual and social lives. That is, Wright tries to
clarify what it is for us—the community of language users—to treat a stretch
of discourse as objective, as we attempt to negotiate and understand this world
we find ourselves occupying.
As Wright sees things, objectivity is not a univocal notion. There are
different notions or axes of objectivity, and a given chunk of discourse can
exhibit some of these and not others. The axes are labelled ‘epistemic con-
straint’, ‘cognitive command’, ‘the Euthyphro contrast’ and ‘the width of
cosmological role’. In a previous paper, Shapiro (2007), I argue that with the
possible exception of some troubling matters near the foundation, mathematics
easily passes all four tests. Mathematics is epistemically unconstrained: there
are unknowable truths. The Galileo observation attests to the extremely wide
cosmological role of mathematics: it figures in all sorts of explanations, most
of which are explanations of non-mathematical matters. Mathematics falls on
the Socrates side of the Euthyphro contrast—it is not response-dependent—
and mathematics easily satisfies cognitive command. In short, mathematics is
objective, if anything is.
On the other hand, the possible exceptions—the foundational matters—
loom large, since they go the heart of the Kant–Quine matter noted just above.
Here I want to revisit one of the axes, cognitive command, since it bears out
this theme, even more than I anticipated in the other paper.
According to Wright, cognitive command figures as an axis of objectivity
only if the discourse is epistemically constrained. So I will pause for a brief
sketch of that primary axis of objectivity.
Epistemic constraint is an articulation of Michael Dummett’s notion of
anti-realism. According to one of Wright’s articulations of this axis (1992,
p. 75), a discourse is epistemically constrained if, for each sentence P in the
discourse,
P ↔ P may be known.
In other words, a discourse exhibits epistemic constraint if it contains no
unknowable truths.
It seems to follow from the very meaning of the word ‘objective’ that
if epistemic constraint fails for a given area of discourse—if there are
propositions in that area whose truth cannot become known—then that dis-
course can only have a realist, objective interpretation:
To conceive that our understanding of statements in a certain discourse is
fixed . . . by assigning them conditions of potentially evidence-transcendent
truth is to grant that, if the world co-operates, the truth or falsity of any such
statement may be settled beyond our ken. So . . . we are forced to recognise
a distinction between the kind of state of affairs which makes such a state-
ment acceptable, in light of whatever standards inform our practice of the
discourse to which it belongs, and what makes it actually true. The truth of
such a statement is bestowed on it independently of any standard we do or can
apply . . . Realism in Dummett’s sense is thus one way of laying the essential
groundwork for the idea that our thought aspires to reflect a reality whose
character is entirely independent of us and our cognitive operations.
Wright (1992, p. 4)
In other words, if epistemic constraint fails for a given discourse, then it

is objective, and that is the end of the story. The other axes of objectivity—
cognitive command, cosmological role and the Euthyphro contrast—are irrel-
evant; they do not apply. On the other hand, if a discourse is epistemically
constrained—if all truths are knowable there—then the other axes track impor-
tant aspects of objectivity. Or so Wright argues. For present purposes, then,
let us just assume that, in mathematics, all truths are knowable, in some
relevant sense of ‘knowable’, and turn to a brief characterization of cognitive
command.
Assume that a given area of discourse serves to describe mind-independent
features of a mind-independent world, understood intuitively. Suppose that two
people disagree about something in that area. It follows that at least one of
them has misrepresented reality, and so something went wrong in his or her
appraisal of the matter. Suppose, for example, that two people are arguing
whether there are seven, as opposed to eight, spruce trees in a given yard.
Assuming that there is no vagueness concerning what counts as a spruce tree
and no vagueness concerning the boundaries of the yard,1 it follows that at
least one of the disputants has made a mistake: she either did not look carefully
enough, her eyesight was faulty, she did not know what a spruce tree is, she
made a mistaken inference, she counted wrong, or something else along those
lines. That is, the very fact that there is a dispute suggests that one of the
disputants has what may be called a cognitive shortcoming (even if it is not
always easy to figure out which one of them it is).
In contrast, two people can disagree over the cuteness of a given baby or the
humour in a given story without either of them having a cognitive shortcoming.
One of them may have a warped sense of taste or humour, or perhaps no
sense of taste or humour, but there need be nothing wrong with his cognitive
faculties. He can perceive and reason as well as anybody.
The present axis of objectivity turns on this distinction, on whether there
can be blameless disagreement. Wright (1992, p. 92) writes that
1 We will return to vagueness later.

A discourse exhibits Cognitive Command if and only if it is a priori that differ-
ences of opinion arising within it can be satisfactorily explained only in terms
of “divergent input”, that is, the disputants working on the basis of different
information (and hence guilty of ignorance or error . . . ), or “unsuitable
conditions” (resulting in inattention or distraction and so in inferential error,
or oversight of data, and so on), or “malfunction” (for example, prejudicial
assessment of data . . . or dogma, or failings in other categories . . .
Intuitively, cognitive command holds for discourse about spruce trees and it
fails for discourse about the cuteness of babies and the humour of stories.
What of mathematics? Wright takes it as obvious that cognitive command
holds for simple calculations (p. 148). Suppose, for example, that two people
differ on the product of two four-digit numbers, after doing the calculation
by hand. Clearly, at least one of them made a mistake. He forgot the relevant
multiplication table (or did not look at it accurately), or else he got the columns
mixed up, or had a lapse in memory or concentration. Any of these clearly
qualifies as a cognitive shortcoming. To fully sanction Wright’s conclusion
here, we would have to deal with Wittgensteinian issues concerning rule-
following—which seems to call the objectivity of that activity into question.
But let us set those issues aside.
In any case, there is a lot more to mathematics than simple calculations.
Does cognitive command hold throughout the enterprize? The epistemic stan-
dard for serious assertion in professional mathematics is proof. So suppose
that one mathematician, Pat, produces what she takes to be a proof of a math-
ematical proposition S; and that another mathematician, Chris, demurs from
S, even after being presented with Pat’s purported proof. The disagreement
between Pat and Chris need not be over the conclusion of the purported proof.
They differ over whether the purported proof is good—whether it establishes
its conclusion. Pat believes S, on the basis of her purported proof, and Chris
does not believe S, as he rejects the correctness of the purported proof. He may
either think that S is false or else he may be agnostic about it.
In these circumstances, our question here is whether we can be sure,
a priori, that at least one of our mathematicians exhibits a cognitive
shortcoming—assuming that the disagreement is genuine, which is another
matter on the table here.
In the world of professional mathematics, disputes like this happen. Two
referees may disagree whether the argument in a submitted article does in fact
prove its conclusion, with the competence of neither referee (nor the author
of the paper) in doubt. But there is nothing special about mathematics here.
There will be similar ‘blameless’ disagreements in any sufficiently complex
area of discourse. To give cognitive command a chance of serving as an
axis of objectivity, and of it helping to shed some light on the status of pure
mathematics, we have to idealize on the cognizers.2
2 I noted above that, according to Wright, cognitive command is an axis of objectiv-

ity only if the discourse is epistemically constrained—only if there are no unknowable
The idealizations in question here are familiar. We assume that our subjects
have unlimited lifetimes, materials, memory and attention spans. These are the
same idealizations invoked in the mathematical theory of computability, and
in mathematical logic generally. Our discussion now threatens to turn back on
itself. To assess whether cognitive command holds in mathematics, and thus
whether mathematics is objective on this axis, we turn to some mathematics,
to negotiate the idealizations. Well, I guess we all know that philosophy is a
holistic enterprise.
For what it is worth, the pursuit of mathematics seems to me more like
discovery than invention, more like getting at the truth than expressing an
attitude. It also seems to me, as an interested outsider, that for the most part, the
mathematical community shows a remarkable tendency toward convergence,
perhaps more so than in any other area. At least at this point in history, dis-
putes concerning the correctness of a given argumentation do not last forever.
Unless the parties simply lose interest—surely a non-cognitive matter—actual
disagreements seem to get resolved to everyone’s satisfaction. Everyone comes
to agree that a certain step is invalid, for example, or that a certain premise is
suppressed, or that the argument is correct after all. This is surely evidence
that cognitive command holds in the world of professional mathematics, at
least when things are suitably idealized.
Define a formal proof to be a sequence of sentences, in a formal language,
in which every step is either an explicitly noted axiom, premise or assumption,
or else follows from previous steps via what, for the author at least, is a
primitive rule of inference, one so basic that there is no sense in breaking it
down further. In actual mathematics, it is not always clear whether a given
proof has a unique formalization. Is there an objective fact of the matter,
independent of judgements and the like, concerning how to formalize any given
piece of published mathematics? Let me concede that an opponent of objectiv-
ity has some room to manoeuver here. The move from actual mathematical
discourse—which is what we care about after all—to the fully formalized
proofs of the idealized mathematicians may not be governed by fully objective
standards. But, as noted above, full objectivity is not likely to be on the cards
in any case. We are exploring the extent to which mathematics is objective, on
this one axis.
Let us return to our imaginary mathematicians Pat and Chris, now suitably
idealized. Assume that Pat puts forward a fully formalized proof of a
mathematical proposition S and that Chris rejects the proof, demurring from
its conclusion.
truths, and I asked us to assume that mathematics is epistemically constrained. To give

that a chance, we also have to focus on idealized mathematicians. There are surely some
true propositions that no flesh and blood mathematician can ever know, simply because
the proposition requires a calculation that is too long to be completed before the Sun
goes cold.
Pat and Chris presumably agree on what formula appears on each line
of the purported proof . A disagreement there would involve a cognitive
shortcoming on the part of one of them; he or she does not have good eyesight.
So either Pat and Chris disagree over one of the axioms, assumptions or
premises of , or they disagree over the validity of one of Pat’s primitive rules
of inference.
In mathematics, a difference over a premise or axiom is, prima facie,
not really a disagreement. The two mathematicians are just talking past one
another. Pat is working in a certain structure (or type of structure), character-
ized in part by the premises of the derivation , while Chris prefers to work
in a different structure. A mathematician who demurs from the Pythagorean
theorem, because he does not assume the parallel postulate, is not in real
disagreement with a Euclidean who does. The two of them work in different
theories, with different subject matters.
Of course, mathematicians did not always think this way. Supposedly,
they once saw the issue concerning geometry as concerning the structure of
(physical) space or intuitions concerning perception, or something else along
these lines. Alberto Coffa (1986, p. 8) describes the historical transition:
During the second half of the nineteenth century, through a process still
awaiting explanation, the community of geometers reached the conclusion
that all geometries were here to stay . . . [T]his had all the appearance of
being the first time that a community of scientists had agreed to accept in
a not-merely-provisory way all the members of a set of mutually inconsis-
tent theories about a certain domain . . . It was now up to philosophers . . . to
make epistemological sense of the mathematicians’ attitude toward geome-
try . . . The challenge was a difficult test for philosophers, a test which (sad to
say) they all failed . . . .
I take it that on the present scene, if Pat and Chris differ only over premises or
axioms, then they do not disagree at all. In effect, their words mean different
things. The two of them simply work in different structures. This explains why
mathematical theories are not discarded as false when they become unusable in
science, at least not now. Michael Resnik (1997, p. 131) calls the phenomenon
‘Euclidean rescue’.
Things may not be this neat if the disagreement concerns a more foun-
dational matter. Suppose that the ultimate conclusion of the ‘disputed’ proof
is a proposition of real analysis, but that Pat’s proof invokes a set-theoretic
principle, such as the continuum hypothesis or a large cardinal hypothesis, and
Chris rejects that set-theoretic principle. This would naturally focus the dispute
on the background set theory. One can invoke a Euclidean rescue there as
well, and say that Pat and Chris work in different structures, just because their
background set theories differ. Pat works in analysis-plus-set-theory-A, while
Chris prefers analysis-plus-set-theory-B. This sort of resolution is not quite as
comfortable as it was with, say, Euclidean and non-Euclidean geometries, due
to the pervasiveness of set-theoretic notions throughout mathematics, and the
foundational role of set theory (see Maddy, 2007, pp. 358–360). The matter can
be murky, and it goes beyond the scope of this paper. We’d have to consider
whether mathematics can have more than one foundation, and, if it does, how
we’d study the relationships between the foundations.
In any case, a dispute concerning the set-theoretic background is much
like our one remaining possibility, that the difference between our mathemati-
cians’ attitudes traces to their logic. Suppose that Chris demurs from a rule of
inference that, for Pat, is so primitive that it cannot be broken down further.
To focus on an example, let us suppose that Pat’s proof invokes instances
of excluded middle, and Chris rejects that, since he is an intuitionist. This
raises the question of the objectivity of logic, which could (and did) take up
another lengthy paper (Shapiro, 2000). The details go beyond present concern,
but I suggest that with some qualifications, logic, too, passes the letter of
all of Wright’s tests for objectivity, with the possible exception of epistemic
constraint. But it is not clear what one should conclude from that.
One potentially troubling matter here there is that most of Wright’s axes of
objectivity are formulated in terms of logic, and so it seems that one can deploy
the various axes only after one has settled on a logic. That is, the various axes
of objectivity presuppose a logic (although it is left open just which logic is
presupposed). So it hard to see the extent to which we can even ask if logic is
objective, using Wright’s framework.
One might invoke Euclidean rescue with logic, taking an eclectic attitude
toward it. The thesis would be that, say, classical analysis and intuitionistic
analysis are two different subjects, and are no more in conflict with each other
than Euclidean and non-Euclidean geometries. If we go down this route, then in
a sense all of mathematics—once suitably idealized—reduces to calculation. It
is just a matter of what conclusions can be drawn in various deductive systems.
There is no content to mathematics beyond that. We save cognitive command,
but at the cost of there not being any interesting, genuine disputes among
mathematicians. The supposedly ‘disputing’ parties do not speak the same
language. We seem to have manoeuvered ourselves into some sort of formalism
concerning mathematics—at least once it is suitably idealized so that we can
apply the axes of objectivity. Of course, we are still putting Wittgensteinian
issues of rule-following aside.
Let us briefly explore the alternative position, that the classical mathemati-
cian (Chris) and the intuitionist (Pat) do have a genuine disagreement with
each other. And then we can ask if the matter is objective, along the axis of
cognitive command. More murky waters lie here.
Our question concerns whether one of our idealized disputants, Chris or
Pat, exhibits a cognitive shortcoming. ‘Inferential error’ appears in the list of
shortcomings that Wright gives in the above characterization of cognitive com-
mand.3 Surely Pat accuses Chris of inferential error. Chris invokes excluded
3 Recall: ‘A discourse exhibits Cognitive Command if and only if it is a priori that

differences of opinion arising within it can be satisfactorily explained only in terms of
middle, and that, for Pat, is an error. However, Wright does not seem to have
this in mind. He takes ‘inferential error’ to be result of ‘inattention or distrac-
tion’, not a deep disagreement concerning the nature of logical consequence
itself. It is part of the idealizations invoked here that neither Chris nor Pat has a
failure of concentration. So if cognitive command is to hold here, we will have
to find some other sort of cognitive shortcoming to attribute to one or the other
of our idealized mathematicians.
Arguably, logic-choice is a holistic enterprise, although this is yet another
controversial matter that cannot be addressed here fully. Resnik proposes an
adaption of the programme of ‘wide reflective equilibrium’ formulated by
Nelson Goodman and used by John Rawls for an account of justice:
One starts with one’s own intuitions concerning logical correctness (or log-
ical necessity). These usually take the form of a set of test cases: arguments
that one accepts or rejects, statements that one takes to be logically neces-
sary, inconsistent, or equivalent . . . One then tries to build a logical theory
whose pronouncements accord with one’s initial considered judgements. It
is unlikely that initial attempts will produce an exact fit between the theory
and the “data” . . . Sometimes . . . one will yield one’s logical intuitions to
powerful or elegant systematic considerations. In short, “theory” will lead
one to reject the “data”. Moreover, in deciding what must give, not only
should one consider the merits of the logical theory per se . . . but one should
also consider how the theory and one’s intuitions cohere with one’s other
beliefs and commitments, including philosophical ones. When the theory
rejects no example one is determined to preserve and countenances none one
is determined to reject, then the theory and its terminal set of considered
judgements are in . . . wide reflective equilibrium.
Resnik (1997, p. 159)
There is another troubling circularity here. In order to see if one is in

reflective equilibrium, she must reason logically. She must draw logical conse-
quences of her logical theory, to see if it coheres with her intuitions and other
‘data’. There can be no rigorous characterization of reflective equilibrium that
is neutral on the choice of logic. We can only speak of reflective equilibrium
for a given logic.
Still, perhaps one of our disputants may fail to be in reflective equilibrium,
by his or her own lights. That would surely count as a cognitive shortcoming
on the part of that disputant. But can we be sure that this always happens? That
is, can we rule out cognitively blameless disagreement?
“divergent input”, that is, the disputants working on the basis of different information
(and hence guilty of ignorance or error. . . ), or “unsuitable conditions” (resulting in
inattention or distraction and so in inferential error, or oversight of data, and so on), or
“malfunction” (for example, prejudicial assessment of data . . . or dogma, or failings in
other categories . . . ’ (Wright, 1992, p. 92)
It seems not, although it is hard to see how one might construct an argument
for this. It seems to be possible that our two idealized mathematicians, Pat and
Chris, are both in reflective equilibrium, each by the lights of his or her own
logic. If so, how can we neutral observers accuse one of them of cognitive
shortcoming?
There is something troubling about the whole issue concerning the objec-
tivity of logic. Any serious dispute in any area of discourse is going to involve
logic. All disputants, in all areas, are themselves reasoners, and come to their
respective conclusions in part by drawing inferences. Given how pervasive
logic is, disagreements or differences about logic are certain to result in dis-
agreements or differences everywhere. If logic fails to be objective, can there
be any objectivity anywhere?
I apologize for failing to come to a crisp conclusion concerning the objec-
tivity of logic and, to that extent, the objectivity of mathematics. Given the
central role of logic in our theorizing, it is hard to separate it out for sharp
treatment. Any attempt to characterize how the question of objectivity is to be
adjudicated will presuppose logic.
The situation can be made more palatable if we recall the Kant–
Quine thesis, the idea that there is no way to sharply separate the parts of our
best theories that are due to the way the world is and the parts that are due
to the way that we, the human cognitive agents, are. One fallout, I submit, is
that there is no sense to asking for complete objectivity. So some feature that
compromises the objectivity of a given area of discourse does not eliminate
objectivity altogether. This is the nature of the holistic beast.
Later in his book, Wright (1992, p. 144) adds some qualifications to the
formulation of cognitive command. It looks like the qualifications are meant
to deal with matters like holistic adjudication, in our struggle for reflective
equilibrium. Wright says that a discourse exerts cognitive command if and
only if
it is a priori that differences of opinion formulated within the discourse,
unless excusable as a result of vagueness in a disputed statement, or in the
standards of acceptability, or variation in personal evidence thresholds, so to
speak, will involve something which may properly be described as a cognitive
shortcoming.
That is, Wright holds that blameless disagreement that turns on vagueness,
standards of acceptability and the like, does not undermine cognitive com-
mand. Why are there these exceptions? What happened to the original motiva-
tion for the criterion of cognitive command, which did not mention vagueness,
evidence thresholds, or the like, one way or the other? Is this an instance of
what Imre Lakatos (1976) calls ‘monster-barring’? We find some parts of our
theory that do not seem to fit, and so we just exclude them.
Although Wright does not put it this way, I suggest that the excep-
tions listed in the nuanced version of cognitive command are in line
with the Kant–Quine theme. On Wright’s views of vagueness—and mine
(Shapiro, 2007a)—vague terms are response- or judgement-dependent, at least
in their borderline regions. He writes that it is ‘tempting to say . . . that a
statement’s possessing (one kind of) vagueness just consists in the fact that,
under certain circumstances, cognitively lucid, fully informed and properly
functioning subjects may faultlessly differ about it.’ Yet, robustly objective
areas of inquiry, such as natural science, are conducted with vague terms.
That, alone, cannot undermine their objectivity, unless one takes an all-or-
nothing approach to that question. Similarly, it is plausible that ‘standards
of acceptability’ and especially ‘variation in personal evidence thresholds’
lie closer to the ‘human’ side and further from the ‘world’ side of the mix.
Surely, a conservative scientist, one who is more cautious in putting forward
claims, need not have any cognitive shortcoming with respect to a slightly more
speculative colleague, nor vice versa. So a disagreement that is traced to that
difference need not undermine cognitive command.
The same may go for the more foundational matters concerning mathe-
matics and its logic. Disagreements that turn on holistic considerations may
end up being adjudicated, in part, by matters of taste concerning, for example,
what a given theorist finds to be elegant or simple. One mathematician may
prefer the subtle distinctions and sharper bounds produced by constructive
mathematics while another might go for the unity and, in some way, simplicity
and tractable meta-theory of classical mathematics. That is, it just may be
that certain foundational matters are negotiated closer to the ‘human’ than the
‘world’ side of the web. Even if this is so, it does not follow that mathematics
is not objective, even a paradigm of objectivity, one of the standards by which
we measure other discourses.
Comment on Stewart Shapiro’s
‘Mathematics and objectivity’
Gideon Rosen
Stewart Shapiro approaches the question of the objectivity of mathematics

indirectly. Instead of asking whether the mathematical facts somehow depend
on human thinking, Shapiro asks whether mathematical discourse exhibits
what Wright calls cognitive command. His chapter presupposes that these ideas
are connected in a simple way:
If a region of discourse fails to exhibit cognitive command, then the facts
with which that discourse is concerned are not objective facts.
In this note I raise a doubt about this principle.
Suppose that you and I disagree about the merits of Wagner’s Ring. You
say it’s a masterpiece; I say it’s a bore. Suppose we both know the operas and
the relevant musicological background very well, that neither of us is drunk
or distracted, that our judgements are stable upon reflection, etc. Given all of
this, we have a disagreement that may persist even though both parties have a
firm grip on the underlying facts and neither has made a mistake in reasoning.
Pressed to explain the disagreement, we may be reduced to saying that we
disagree simply because we bring incompatible but equally coherent aesthetic
sensibilities to the question.
An area exhibits cognitive command, roughly speaking, when every dis-
agreement is the result either of divergent input—i.e., differences in the infor-
mation available to the disputants—or some sort of cognitive malfunction or
mistake in reasoning. The example suggests that aesthetic discourse fails to
exhibit cognitive command and hence, given the above presupposition, that
aesthetic facts are not objective facts. Shapiro argues (with some qualifications)
that mathematical discourse passes the test with flying colours, and hence that
so far as this criterion is concerned, there is no reason to doubt the objectivity
of mathematics.
Is this right? Modern mathematics is ultimately a matter of proving theo-
rems from axioms by means of rules. Any mathematical disagreement that is
not due to a simple mistake is therefore traceable either to disagreement about
the axioms or to disagreement about the rules of logic. Let us set disagreement
about logic to one side. This is an important issue, but as Shapiro notes, it is
hard to discuss within Wright’s framework. If we focus on disagreement about
the axioms, then the first thing to emphasize (as Shapiro does) is that while
disagreement of this sort has occasionally surfaced in mathematics, e.g., in
the debate over the parallel postulate in geometry, modern mathematics has a
standard way of finessing it. The idea is to treat the disputed axioms as clauses
in the definitions of special mathematical structures. Where geometers might
once have disagreed about the absolute truth or falsity of the parallel postulate,
a modern mathematician will say: ‘Some spaces are Euclidean, others aren’t.
The axiom holds in every Euclidean space, since conformity to the axiom is
part of what makes a space Euclidean. But it fails in other spaces, examples of
which are easily given’. On this account, it makes no sense to ask whether the
axiom is true or false simpliciter, and so it makes no sense to disagree about its
truth.
Apparent conflicts in mathematics can usually be dissolved in this way—
but not always. One central concern of mathematics is to establish the existence
of mathematical objects satisfying various conditions. Now an existence proof
always requires at least one existence axiom, and if one examines real math-
ematics one finds that it is always legitimate simply to assert the existence of
the natural numbers and of certain sets of natural numbers. These existential
claims are not mere conditions or assumptions. When a mathematician proves
the consistency of hyperbolic geometry by producing a model of the axioms
in (say) R3 , the content of his theorem is not ‘If there are numbers, etc, then a
model of the axioms exists.’ His proof establishes the existence of a model
outright, and this means that it must involve the assertion of at least one
existential axiom.
There is of course no real disagreement within mathematics about the
existence of the natural numbers and certain sets constructed from them. But
in applying the cognitive command test we are not constrained to focus only
on real disagreement. Even if there were no actual disagreement about the
merits of the Ring (thanks to uniform musical education), the mere possibility
of the disagreement described above would suffice to show that aesthetic
discourse flunks the cognitive command test for objectivity. By the same token,
the mere possibility of disagreement about the existential claims of standard
mathematics would suffice to establish the non-objectivity of at least part of
mathematics, provided that disagreement was not traceable to ‘divergent input’
or ‘mistakes in reasoning’.
Such disagreement is clearly possible. Mathematicians who have been
acculturated in the normal way find the existential claims of basic arithmetic
perfectly obvious and hence acceptable without proof. But we know that it
is possible for a person to find these axioms anything but obvious. After all,
M AT H E M AT I C S A N D O B J E C T I V I T Y: C O M M E N T 111
some philosophers explicitly reject them on the grounds that (a) they have
no intrinsic plausibility, and (b) every positive argument in their favour is
unconvincing (Field, 1980; Leng, this volume). These philosophers typically
point out, for example, that numbers would be invisible, non-physical entities
of some sort, and that it is hardly evident to them that such things exist.
I want to imagine this as a genuine clash of sensibilities. Some people find
the existential axioms obvious and so affirm them; others find them totally non-
obvious and withhold assent. Is this a matter of ‘divergent input’? Of course
the axioms strike them differently. But if we treat this as a matter of divergent
input in applying the cognitive command criterion, we shall have to say that
our disagreement about Wagner is also a matter of divergent input, and that
would trivialize the criterion. Is the disagreement due to a mistake in reasoning,
or to some other cognitive malfunction? Perhaps; but if it comes down to a
disagreement about the intrinsic plausibility of an axiom, it is hard to see why
this should be so.
The possibility of a disagreement about the existential assumptions of stan-
dard mathematics suggests ordinary mathematics may not exhibit cognitive
command. Should we conclude that mathematics is not objective after all? No.
We should conclude instead that cognitive command is a flawed criterion of
objectivity. Objectivity in the relevant sense is a metaphysical concept. To call a
fact objective is to say that it does not depend in any interesting way on thought
or language (Rosen, 1994). From the fact that a disagreement in mathematics
might be traceable to variation in our sense of ‘mathematical plausibility’,
nothing follows about the metaphysical status of its subject matter. If we
discovered that disagreement about the existence of God were sometimes
traceable to a difference in theological ‘sensibility’, would we conclude that
God’s existence (or non-existence) was somehow mind-dependent? We might
conclude that judgements about God’s existence are not strictly forced upon
us by the evidence. But this is a point about the epistemic status of these
judgements; it is not a point about the metaphysical status of the facts with
which they are concerned.
Reply to Gideon Rosen
Stewart Shapiro
I tend to agree with the main thrust of Gideon Rosen’s comment on my

‘Mathematics and objectivity’. In particular, I agree that Crispin Wright’s
notion of cognitive command does not, by itself, provide a necessary and
sufficient condition for objectivity. The purpose of my paper (and two others) is
to test mathematics and Wright’s various axes of objectivity against each other.
However, I do think that there is something right about cognitive command as
at least a defeasible criterion; it is a matter of articulating its scope and limits.
I hold, with Wright, that objectivity is not a univocal notion. There are various
aspects of objectivity, not all of which line up with each other.
In another paper, ‘Objectivity, explanation, and cognitive shortfall’ (forth-
coming in a festschrift for Crispin Wright), I provide a thought experiment
involving two scientists who disagree with each other, but are each in reflective
equilibrium concerning the overall balance of evidence. It would be hard to
fault either of them. Yet one would be loath to conclude that science generally
is not objective. A referee for that paper pointed out that Wright’s criterion,
by itself, does not distinguish failures of cognitive command due to the non-
objective nature of the subject matter (e.g., Rosen’s example about Wagner’s
Ring) and failures due to the scantness of evidence, especially in such areas
where evidence is evaluated holistically. Perhaps Rosen’s examples concerning
mathematical existence, and the general philosophical themes of fictionalism,
can be understood similarly. The referee located this in the history of other
failed attempts to demarcate cognitive significance.
Toward the end of his commentary, Rosen suggests that cognitive com-
mand looks in the wrong place: objectivity is a purely metaphysical matter,
while cognitive command is a broadly epistemic criterion. I do not agree with
that perspective, but this is not the place to pursue that general issue.
9
The reality of mathematical objects
Gideon Rosen
If the truth be known, there are no such things as numbers; which is

not to say that there are not at least two prime numbers between 15
and 20.
Paul Benacerraf, ‘What numbers could not be’ (1965)
The problem
The closing sentence of Paul Benacerraf’s famous paper is a kōan: a bit of
seeming nonsense that points—or seems to point—to a deep truth. It is a
theorem of basic arithmetic that there are two prime numbers between 15 and
20. Anyone who accepts basic arithmetic must therefore agree that there are
two prime numbers between 15 and 20, and hence there are at least two num-
bers, and hence that there are numbers. And yet the idea that numbers are real
things—that the real world contains mathematical objects in the same sense in
which it contains guns and rabbits—can sound preposterous or confused. And
so we find philosophers straining to articulate a position of the following sort:
Of course there are numbers (and functions and sets and mathematical
objects of other sorts). That’s just mathematics, and we have no quarrel with
mathematics. But in another sense—the metaphysical sense—there are no
numbers. Numbers are not real. Numbers are not Things.1
It should be obvious that such remarks are seriously puzzling as they stand.
Suppose your exterminator tells you that you have squirrels in your attic,
but then goes on to add that in the strict and philosophical sense there are
no squirrels. Or suppose an astrophysicist reports that there are three black
1 For a recent statement of this idea, see Dorr (2008).

holes at the center of the galaxy, and then goes on to say, ‘And by the way,
black holes are not real; they are not Things.’ If you are anything like me,
you will find these remarks incomprehensible. And yet in the philosophy of
mathematics there is a widespread sense that such formulations, while perhaps
less transparent than one might like, must nonetheless make some sort of
sense.
Let us call the thesis we seek to clarify qualified realism about mathemat-
ics. The view is a form of realism because it holds that there is a sense in
which mathematical objects exist. It is a qualified realism because it holds that
these mathematical objects are somehow metaphysically ‘second rate’. The
difficulty is to say what this qualification comes to.
I used to think that this could not be done in such a way as to yield a version
of qualified realism worth discussing (Rosen and Burgess, 2005; Rosen, 2006).
I now suspect that this pessimism was premature. My aim in this note is
to present an account of what the qualified realist might have in mind—an
account according to which the objects of mathematics compare unfavourably
in certain metaphysical respects with certain paradigmatic Things: the objects
of everyday experience, perhaps, or the objects of the physical sciences. I
do not endorse the version of qualified realism that I shall discuss, but I
believe it merits our attention. The aim here is simply to put the view on the
table.
The case for minimal realism about mathematical objects

Before we begin, it may help to review the case for realism itself (Burgess,
1983; Burgess and Rosen, 1997; Rosen and Burgess, 2005).
By minimal realism about mathematical objects, I mean the claim that
mathematical objects exist. I use the phrase ‘mathematical object’ as a catch-
all to cover numbers, functions, sets, groups, spaces, models, vectors, cate-
gories, systems of equations, formal languages, and the other manifold items
with which mathematics is distinctively concerned. To say that mathematical
objects exist is just to say that there is at least one item of this sort, or
equivalently, that one such thing exists.
Must I say what I mean by the word ‘exist’ in this context? I don’t think
so. The existential idioms—the predicate ‘exists’, along with quantificational
expressions like ‘there are . . . ’ ‘there exist . . . ’, ‘at least one . . . ’, etc.—are
part of the everyday language of mathematics. These idioms are all equivalent
in that language, and they are not ambiguous. That is why they can all be
represented by a single symbol, ∃, in standard formalizations of mathematics.
If you are reading this paper, you already understand that language, and I
propose to rely on that understanding. So when I say that minimal realism
is the thesis that mathematical objects exist, I might add that I mean to use the
word ‘exist’ in its ordinary mathematical sense, the sense a high-school student
THE R E A L I T Y O F M AT H E M AT I C A L O B J E C T S 115
or a professional mathematician has in mind when he says that there exist two
solutions to some equation. This is not a definition, but it should suffice.2
So understood, minimal realism is not an esoteric metaphysical claim. It
is a mathematical claim—a trivial consequence of the most elementary parts
of mathematics. As noted, it is a theorem of arithmetic that there are two
prime numbers between 15 and 20. This theorem entails that there are at least
two numbers, and hence two mathematical objects. Anyone who accepts basic
arithmetic therefore has no choice but to accept minimal realism.
Should we accept basic arithmetic? Are we justified in believing that
there are two prime numbers between 15 and 20? In my view, the claims of
elementary mathematics have the same status as the claims of common sense
in other areas—e.g., the claim that we inhabit a world of real things that exist
even when we are not aware of them, or the claim that other human beings
have conscious mental lives rather like our own. Common sense is fallible, of
course. But if a philosopher (or a scientist or anyone else) wishes to call the
claims of common sense into question, he must give reasons for his doubts.
And when it comes to basic arithmetic, such reasons simply do not exist.
Arithmetic is obviously unassailable on mathematical grounds.3 And if we
treat ordinary scientific standards for the acceptance and rejection of theories
as authoritative, the fact that every developed science takes mathematics utterly
for granted is enough to show that the scientific enterprise broadly conceived
has never thrown up doubts about arithmetic. If there are any grounds for doubt
in this area, they must therefore be distinctively philosophical grounds.
I will not try to survey the arguments that philosophers have developed in
this context, but I will say (very briefly) why I find them unpersuasive. The
existing arguments are of two kinds. Some philosophers say that we should
reject standard mathematics because numbers and the like would be queer
things if they existed. And this is certainly true. If there is such a thing as the
number 26, it is very different from the objects of everyday experience (tables,
etc.) and the less-familiar objects disclosed by physics (quarks, etc.). But so
what? To advocate the rejection of arithmetic on this ground is repose more
confidence in a grand metaphysical scheme (sometimes called physicalism)
according to which absolutely everything is rather like a table or a quark,
2 I might simply have said that I use ‘exist’ and the other existential idioms to mean
what they mean in ordinary English. As Quine stressed, there is no good reason to
believe that these words mean one thing in mathematics and something else in other
areas (Quine, 1960, §27). Numbers are plainly very different from tables and quarks
and mental images. But if all of these things exist—i.e., if there are such things—there
is no reason to suppose that they exist in different senses.
3 Some philosophically minded mathematicians have expressed doubts about certain
parts of classical arithmetic—the impredicative or non-constructive parts (for example,
Nelson, 1986). These specialized disputes need not concern us. Constructive arithmetic
involves existence claims, e.g., the claim that there are two prime numbers between 15
and 20.
than in the arithmetical claim that there are two prime numbers between 15
and 20. But if it is a matter of choosing between grand metaphysical schemes
and basic arithmetic, it is clear to me that it is the metaphysics that ought to
budge.
Other philosophers reject basic arithmetic because they think that if num-
bers and other mathematical objects existed, there would be no way for us
to know anything about them. This sort of claim is typically supported by
a general philosophical theory of knowledge, according to which knowledge
requires some sort of interaction between the would-be knower and the object
of his inquiry (Benacerraf, 1973). These theories were originally developed
to account for empirical knowledge, and they may be useful for that purpose.
But if the philosopher insists that they hold in complete generality, he faces a
glaring difficulty. These restrictive theories typically entail that the usual ways
of fixing opinion in mathematics—calculation, proof, informal mathematical
argument—could not possibly be sources of knowledge (since they do not
involve causal intercourse with the numbers), and hence that someone who has
come to believe in the ordinary way that 235 + 657 = 892 does not really know
that 235 + 657 = 892. But then the question arises: Why isn’t this simply a
counter-example to the philosopher’s theory? In other areas, when a philosoph-
ical theory is incompatible with an otherwise uncontroversial fact, the normal
response is to reconsider the theory or to restrict its scope. A philosophical
theory of knowledge must accommodate the manifest fact of mathematical
knowledge. If the theory clashes with this fact, so much the worse for the
theory.
This is just a rough sketch of an intricate dialectic, but the main tactic
should be clear. The central claims of mathematics are unassailable by all
pertinent mathematical, scientific and commonsensical standards. Any philo-
sophical challenge to these claims is thus a sceptical challenge of a certain
special sort, one that depends on bringing distinctively philosophical principles
to bear on questions that are not themselves purely philosophical. Sceptical
challenges of this sort are notoriously weak. When speculative philosophy
contradicts settled science or common sense, the normal response—and, I
believe, the reasonable response—is to suspect that it is the philosopher who
is mistaken. This is not a firm principle, but it is a good rule of thumb. And if
we follow it in this case we have no choice but to allow that since there are two
prime numbers between 15 and 20, mathematical objects therefore exist.
Qualified realism: an example

Minimal realism, then, is not just a philosophical claim; it is a (trivial) bit of
settled mathematics. And yet for many philosophers the view conjures up a
repugnant picture according to which mathematics is like zoology: a science
whose aim is to describe the curious behaviour of a special class of things,
the main difference being that in the mathematical case the objects of interest
are infinite in number and totally invisible.4
It is this thought, among others,
that has led philosophers to grope in the direction of the view we have called
qualified realism: the view that numbers are not really things in the sense
in which lions and tigers are things. Our task is to give a sense to this dark
thought.
We begin with some examples of views that seem to encourage this sort of
verbiage. The examples are all versions of reductionism about arithmetic. In
general, a reductionist thesis in this area holds that the arithmetical facts are
somehow grounded in or amount to nothing over and above facts of another,
more fundamental, kind. (These italicized phrases raise an eyebrow, and we
shall return to their interpretation.) Now some reductionist proposals of this
sort have no interesting metaphysical implications. If a theorist identifies the
individual numbers with pure sets—say, by claiming that 0 is the empty set,
and that the successor of a number n is the set whose sole member is n—then
every fact of arithmetic may be reduced to a fact of set theory. And yet this sort
of reduction by itself has no tendency to impugn the Reality of the numbers.
After all, it is consistent with this view that sets are as robust and thing-
like anything could be, and since the view identifies numbers with sets, it is
therefore consistent with unqualified, full-strength realism about the numbers.
We get a more interesting form of reductionism when we consider propos-
als that purport to reduce the truths of some part of mathematics to truths in
some more basic idiom without identifying the objects of mathematics with
objects recognized in the more basic theory. As an illustration, consider for-
malism in the philosophy of arithmetic. The formalist’s central thought is that
arithmetic is not ultimately concerned with an extralinguistic domain of things.
Rather, insofar as arithmetic has a proper subject matter, it is the language of
arithmetic itself and certain formal relations among its sentences.5 Here is a
simple version of the view. Let PA be the usual formalization of elementary
arithmetic: first-order Peano arithmetic.6 Let PAω be this theory supplemented
4 ‘Arithmetic as the natural history (mineralogy) of numbers. But who talks like this
about it? Our whole thinking is penetrated with this idea.’ (Wittgenstein, 1956: 1967,
IV, p. 11)
5 The view has few contemporary adherents among philosophers, though mathe-
maticians often find it congenial. See Curry (1951) for one version. For an exposition
and assessment of Frege’s celebrated objections to formalism, see Resnik (1980).
6 The axioms include the basic principles governing the successor function:
0 is a number
0 is not the successor of any number
Every number has a successor
No two numbers have the same successor;
the recursion equations for addition and multiplication;
For any number x, x + 0 = x
For any numbers x and y, x + the successor of y = the successor of (x + y)
with an infinitary rule of inference—an omega rule—that permits the infer-
ence from an infinite list of premises A(0), A(1), . . . A(n) . . . to the universally
quantified conclusion: For all numbers x, A(x). PAω is obviously a sound
theory. The axioms are true, and the rules of inference preserve truth. More
importantly, PAω is also a complete theory, in the sense that every sentence A
in the language of arithmetic is such that either A or its negation is provable in
PAω .7 This means that anyone who accepts standard arithmetic should accept
the following equivalence:
For any sentence A in the language of arithmetic, A is true if and only if A is
provable in PAω .
Taken by itself, this equivalence is a mathematical fact with no special
metaphysical significance. Even the most unreconstructed Platonist should
accept it. But now consider the formalist’s characteristic claim:
For any true arithmetical sentence A:
A is true in virtue of the fact that A is provable in PAω ; or
What makes A true is the fact that it is provable in PAω ; or
A’s truth is grounded in the fact that A is provable in PAω ;
A’s truth consists in the fact that A is provable in PAω .
The italicized idioms here are not part of the official vocabulary of mathemat-
ics. Mathematics assures us that claims about the natural numbers are equiv-
alent to claims about the formal provability of certain sentences in a certain
formal system. But the formalist’s distinctive claim is that the arithmetical
facts are somehow grounded in these proof-theoretic facts, and that the proof-
theoretic facts are therefore, in a corresponding sense, more fundamental. From
the standpoint of mathematics, this claim is extracurricular. It is a distinctively
philosophical claim about the metaphysics of arithmetic.
Note that even though the reductive formalist regards certain linguistic
items (sentences and formal systems) as more fundamental than the numbers,
he cannot deny that numbers exist. Since the existence theorems of arithmetic
(e.g., there are two primes between 15 and 20) are all provable in PAω , the
For any number x, x × 0 = 0

For any numbers x and y, x× the successor of y = (x × y) + x;
and an axiom scheme for mathematical induction:
If 0 is F, and if for all x, F(x) implies F(the successor of x), then every number is F.
7 The completeness of PA is consistent with Gödel’s famous incompleteness the-

ω
orem. Gödel’s theorem shows that any consistent theory that includes basic arithmetic
must be incomplete, provided the theory is recursively enumerable. Roughly speaking,
a theory is recursively enumerable when there exists a finite mechanical procedure—a
computer programme—for listing its theorems. PAω is not recursively enumerable in
this sense, so Gödel’s theorem does not apply.
formalist must concede that these theorems are true. But as we have seen,
if this particular theorem is true, two prime numbers exist. So the formalist
must say (as emphatically as one likes) THERE ARE NUMBERS; NUMBERS
EXIST. What matters for our purposes is what he goes on to say, viz., that all it
takes for numbers to exist is for certain sentences to be derivable from certain
others in a certain formal calculus. It is at this point that he begins to sound
like something less than a full-strength realist about the numbers.
One is tempted to explain this as follows. If the numbers were first-class
things then our claims about them would be true, when they are true, in part
because the numbers are as they are. That’s how it is in zoology. Penguins are
real things, and true claims about them are rendered true, at least in part, by
the birds themselves. In arithmetic as the formalist understands it, by contrast,
a statement about the numbers is true (when it is) because it is derivable
from certain axioms in accordance with certain formal rules. The numbers
themselves play no role in grounding the truths of arithmetic. In fact it is
natural to suppose that it is the other way around. If we ask the formalist to
tell us why Euclid’s theorem on the infinity of the primes is true, he will say
that it is true because it is a theorem of PAω . And if we then ask him why
there are infinitely many prime numbers—that is, if we ask him a question
about the numbers themselves and not about the truth of a certain sentence—
he may say: there are infinitely many prime numbers because the sentence
‘There are infinitely prime numbers’8 is true. On this account (which goes a bit
beyond the views explicitly attributed to the formalist above)9 , not only do the
mathematical objects play no role in making our mathematical theories true,
the objects themselves exist only because our claims about them are true for
other reasons. And as soon as one hears this one has the palpable sense that
formalism amounts to a form of qualified realism about the numbers.
Another example: modal structuralism

As another example, consider a version of structuralism in the foundations
of arithmetic. The structuralist begins from an important insight, namely, that
whenever we prove a theorem about the natural numbers, we in fact prove
a more general theorem about any collection of objects isomorphic to the
numbers. The natural numbers in the standard order
0, 1, 2, . . .
8 . . . or its surrogate in the language of formal arithmetic.

9 To be explicit, we are now imagining a version of formalism that involves two
schematic claims:
For true arithmetical statements A:
(i) The fact that ‘A’ is true is grounded in the fact that ‘A’ is a theorem of PAω ;
(ii) The fact that A is grounded in the fact that ‘A’ is true.
exemplify a distinctive pattern: a discrete linear order with no last element,
in which every item has only finitely many predecessors. If the succession of
Roman emperors had gone on forever, the sequence
Augustus, Tiberias, Gaius,. . .
would have been another instance of this pattern. We call any ordered collec-
tion of this sort an omega sequence. In general, an ordered collection is a pair
(X, ≺) consisting of a set X and a relation ≺ on that set. In the hypothetical
example, the Roman emperors together with the temporal relation x’s reign
precedes y’s reign constitutes an omega sequence, as do the natural numbers
together with the relation x is less than y.
In order to state the relevant fact about arithmetic, we need one more bit
of received wisdom. It is well known that given a suitable logical background,
every arithmetical claim that can be formulated in the elaborate technical lan-
guage of modern number theory can be expressed in a stripped-down language
in which the only primitive symbols are N (for natural number) and < (for
the standard less than relation). If A is an ordinary arithmetical sentence (e.g.,
there are two prime numbers between 15 and 20), let’s call its translation into
this stripped-down language A[N, <].
The pertinent theorem is then as follows:
For any claim A in the language of arithmetic, A[N, <] if and only if, for any
omega sequence (X, ≺), A[X, ≺].
Here A[X, ≺] is the result of replacing the specifically arithmetical vocabulary

in A[N, <]—that is, every occurrence of N and <—with variables X and ≺
ranging over sets and relations on those sets. Thus if A[N, <] is a sentence
in the language of arithmetic that says the natural numbers exhibit a certain
arithmetical feature, the right-hand side of this equivalence says that any
omega sequence—even one whose elements are Roman emperors—exhibits
a corresponding purely structural feature. The theorem thus entails that every
arithmetical claim about the numbers is equivalent to a perfectly general claim
about omega sequences.
Now at this point we must note an assumption of this little theorem. An
omega sequence is an infinite set, so if there are no infinite sets there are no
omega sequences. But if there are no omega sequences, then every instance of
the right-hand side of the equivalence is trivially true. This means that the
equivalence holds only if infinite sets actually exist, and one may wish to
avoid this assumption. One way to do this is to consider a somewhat different
equivalence. Even if there are in fact no infinite collections, there could have
been. The Roman Empire did not endure forever, but it could have. With this
in mind, we might consider the following equivalence.
A[N, <] if and only if as a matter of necessity, for any omega sequence
(X, ≺) A[X, ≺].
Even if there are no infinite sets in the actual world, the claim that any
possible omega sequence has a certain structural feature A(X, ≺) is non-trivial
(on the assumption that omega sequences are possible). The theorem affirms
that this modal claim is equivalent to the ordinary mathematical claim A
with which we began. (A modal claim is a claim about what is possible or
necessary.)
Now so far this is just a bit of uncontroversial (if somewhat unfamiliar)
mathematics. Anyone who accepts standard arithmetic should accept this
equivalence. The structuralist’s characteristic philosophical claim is that the
truths of arithmetic reduce to or are grounded in general claims about all
possible omega systems.10 Return to our example, the claim that there are two
prime numbers between 15 and 20. The structuralist accepts this claim, since
he accepts ordinary arithmetic as it stands. And this means that he (like the
formalist) cannot deny that numbers exist. His distinctive claim is that this fact
about the numbers is grounded in the fact that if there were an omega sequence
of whatever kind, it would have a certain complex structural property—very
roughly, the property of having two ‘prime elements’ in between in its 16th
and 21st elements.11 And here is the crucial point: on the face of it, this latter
fact is not a fact about a distinctive kind of object. It is not a fact specifically
about numbers, nor is it a fact specifically about Roman emperors. Indeed there
is a sense in which it is not a fact about anything at all. After all, a conditional
modal claim of the form if there were an infinite set of a certain kind, it would
exhibit such and such features does not affirm the existence of anything in the
actual world. According to one sort of structuralist, the truths of arithmetic
are made true by modal facts of just this sort. And it is not hard to see why
someone who takes this view might be tempted to say that there is a sense
in which the numbers are not real things. The truths about penguins are made
true by the birds and their behaviour. The truths about the numbers, by contrast,
are not really made true by the numbers; rather they are made true by general
conditional facts in which the numbers themselves make no appearance.
A framework for formulating the proposal

Examples could be multiplied, but the pattern should be clear. Mathematical
objects of a given sort (e.g., the natural numbers) are said to be reducible when
every truth about them obtains in virtue of some truth in a more fundamental
idiom in which the objects in question do not figure. In our examples, the
reducing facts—the facts about provability in PAω , or the facts about every
possible omega sequence—involve no explicit reference to (or quantification
10 This is a version of ‘eliminative’ structuralism modelled on Benacerraf (1965),

Putnam (1967), and Hellman (1989). A different view, also called structuralism, is
defended by Resnik (1997) and Shapiro (1997).
11 This assumes that the natural numbers begin with 0.
over) the numbers, and yet every fact about the numbers is said to obtain in
virtue of some such fact. If a view of this form is correct, then while the
numbers are perfectly real in one sense—there are such things—they may be
‘unreal’ in the sense that when we inspect the facts that ultimately ground
the truths of arithmetic, we find no numbers of any sort. This suggests a
slogan: real things do not disappear upon reduction. One way to reject the
‘mathematical objects picture’, as Hilary Putnam calls it (Putnam, 1967), is to
hold that mathematical objects are unreal in just this sense.
The discussion thus far trades one mystery for another. We began by
wondering what it could mean to say that numbers and other mathematical
objects aren’t really things. The proposal is to explain this notion in terms
of another: the idea that truths of one sort reduce to truths of another sort.
And yet this notion is notoriously problematic. Certainly there has been a
great deal of confusion about what it means to say that truths about (say) our
conscious mental lives reduce to truths about (say) the physiological processes
in our brains and bodies. There are many conceptions of reduction, and hence
many ways to interpret our proposal. Rather than survey the options, I will
simply sketch one conception that strikes me as especially useful for our
purposes.
As I understand it, reduction is a relation among facts—not among sen-
tences or statements, but rather among the facts or states of affairs those
sentences purport to describe. Reduction is thus a metaphysical relation, not
a semantic one. To say that facts of one sort reduce to facts of another sort
is not to make a claim about the meanings of words; it is to make a claim
about the facts themselves, which typically obtain quite independently of our
capacity to speak and think about them.
For present purposes we should think of facts as complex entities built up
from objects, properties, relations and various other items in much the same
sense in which sentences are built from words. For example, the fact that
2 + 3 = 5 is a complex that might be represented as follows:
[= (+(2, 3), 5)]
Here the numbers 2 and 3, along with the identity relation and the operation
of addition are literally constituents of the fact, just as the words ‘2’ and ‘3’
are constituents of the sentence ‘2 + 3 = 5’.12 (In what follows, a sentence
in square brackets names a fact whose structure corresponds to that of the
enclosed sentence.)
My main substantive assumption is that facts stand in a basic relation of
grounding. There is no standard English word for this relation, and there is no
settled bit of philosophical terminology that picks it out. But we have a number
of familiar idioms that point in the right direction, as when we say that one fact
12 For my purposes, facts might be identified with true structured propositions of the
sort originally described by Bertrand Russell (1905).
obtains in virtue of another, or that one fact makes another fact obtain. Some
examples may help:
Disjunctive facts are grounded in their true disjuncts. It is a fact that I am
now either in New Jersey or in Cambridge. The fact obtains (as it happens)
in virtue of the fact that I am in New Jersey. If I had been in Cambridge, the
same fact would have obtained for a different reason.
Existential facts obtain in virtue of their instances. It is a fact that someone
spilled the milk. As it happens, this fact is grounded in the fact that Fred
spilled the milk. If someone else had spilled it, the same fact would have
obtained for a different reason.
Facts about the determinable features of things are grounded in more deter-
minate facts. A certain ball is red in virtue of being (say) crimson; the
particle has a mass of between 10 and 20 MeV in virtue of the fact that
its mass is exactly 17.656 MeV.
Facts involving definable properties and relations are grounded in their
‘definitional expansions’. To be a square is, by definition, to be an equilateral
rectangle (or so we may pretend). Given this, if ABCD is a square, then it is
square in virtue of the fact that it is equilateral and rectangular. That is what
makes the figure in question a square.
Supervenient facts are typically grounded in the facts upon which they
supervene, even if we cannot state the patterns of dependence in a systematic
way. The fact that the US trade deficit with China in 2008 was roughly $117
billion supervenes upon a vast mosaic of particular facts about individual
economic transactions, and perhaps ultimately on some vaster array of facts
about the quarks and electrons that composed the people who did the buying
and selling. Macroeconomic facts obtain in virtue of these lower-level facts,
even if it is impossible in practice, and perhaps even in principle, for us to
spell out the micro facts in virtue of which any given macro fact obtains.
These examples bring out two important features of the grounding relation.
First, it is a form of necessitation. The facts that ground a given fact entail
the fact they ground as a matter of absolute necessity. This distinguishes the
grounding relation from certain forms of causal or nomological determination.
There is no doubt a sense in which effects are grounded in their causes: the
cause makes the effect occur, and so on. But as Hume noted, it is always
possible for the cause to occur without the effect. The grounding relation
that interests us involves a much more intimate form of dependence, as the
examples show.
Second, the grounding relation is an explanatory relation. To cite the facts
that ground a given fact is to give information about why that fact obtains. To
suppose that the grounding relation is an objective relation, as I do, is therefore
to suppose that there are objective facts about the explanatory order (which is
not to say that the practice of giving explanations is always a matter simply of
reporting these objective facts).
These remarks do not amount to a definition of the grounding relation. My
own view is that the relation is too basic to admit of definition. More might be
said by way of informal explication, of course, but I hope that this much will
suffice for our present purposes.13
The proposal
A reductionist proposal in the philosophy of mathematics is a claim to the
effect that every mathematical fact of a certain kind—e.g., every truth of
arithmetic, or every truth of set theory—is ultimately grounded in facts of a
rather different kind, e.g., acts about formal provability, or about what would
have been the case if there had been an infinite sequence of objects.14 We have
considered two examples in which a reductionist proposal of this sort seems to
imply that the objects of arithmetic are somehow less real or less ‘thing-like’
than certain other things. And in both of these proposals the reduction has had
a distinctive character: the objects of the higher-level theory do not figure in the
more fundamental facts to which that theory has been reduced. This suggests
a natural strategy for explaining the metaphysical thesis of qualified realism.
Let us say that a fact is fundamental if it is not grounded in further facts,
and that a thing is fundamental if it is a constituent of a fundamental fact.
Then we might identify full-strength realism about mathematics with the thesis
that some mathematical objects are fundamental things. This is certainly the
view of the hardcore Platonist for whom the numbers are sui generis abstract
substances—invisible luminous spheres arrayed in Plato’s heaven. But it is also
the view of more moderate Platonists who regard arithmetic, or perhaps some
more comprehensive theory like Zermelo-Frankel set theory, as an autonomous
body of truths not grounded in anything more basic. The forms of qualified
13 The material in this section is developed in Rosen (2010).

14 It is worth noting that the reductionist may also make a stronger claim, viz.,
that for every mathematical proposition p, for p to be the case just is for q to be the
case—where q is a proposition involving objects of a more fundamental sort. Consider
a standard scientific reduction: for it to be the case that x is hotter than y just is for it
to be the case that the mean kinetic energy of the particles composing x is greater than
the mean kinetic energy of the particles composing y. We may say that the fact about
temperature strongly reduces to the fact about mean kinetic energy. Strong reductions
of this sort always entail claims of necessary equivalence: if p strongly reduces to q,
then as a matter of necessity, p is true if and only if q is true. That is enough to show
that strong reduction differs from the grounding relation that we have been discussing.
[p or q] may be grounded in [q], but no one would say that for p or q to be the case
just is for q to be the case. In general, [p] may be grounded in [q] even though [p] does
not strongly reduce to [q]; but not vice versa. If [p] strongly reduces to [q], then [p] is
grounded in [q]. In Rosen (2009) this claim is called the Grounding-Reduction Link.
realism that we have discussed are all opposed to mathematical realism of both
these sorts.
Stated more generally, the proposal for explaining the puzzling verbiage
associated with qualified realism is as follows:
Qualified realism about F’s is the thesis that F’s exist, but no fundamental
fact contains an F as a constituent.
When a philosopher tells you that numbers are not real in the metaphysical
sense (even though they exist in the mathematical sense) one thing he or
she might mean is this: that every fact in which a number figures ultimately
obtains in virtue of some collection of facts in which numbers do not figure as
constituents.
The idea gains support from reflection on the ontological status of ‘emer-
gent’ or ‘higher-level’ entities in other areas. Consider the US dollar, or the
European Union or the information encoded on my hard drive. There is a
straightforward sense in which these things exist. The dollar exists as a genuine
form of currency in the sense in which the Italian lira once existed but no
longer does. There is a European Union in the sense in which there is not,
but might one day be, an Intergalactic Federation. My hard drive really does
encode a certain body of information—information that might be destroyed if
I press the wrong sequence of keys. And yet there is a powerful tendency to
think that while these existential claims are all perfectly correct, it would be
a mistake to think of the US dollar as a Thing in the sense in which a table
or God—or a number as conceived by the Platonist—would be a Thing. The
proposal allows us to make sense of this tendency. It is very natural to suppose
that every fact about the dollar or the EU is ultimately grounded in facts about
things of a different sort—facts about the attitudes and actions of economic
actors and the gold reserves in Fort Knox; facts about the legal arrangements
among EU states; etc. Of course there is little hope of stating in any finite
way the complete account of that in virtue of which the dollar is currently
valued at 0.6825 euros. But the proposal does not require that. Unlike some
earlier conceptions of reduction, reduction in our sense is not a thesis about the
meanings of sentences in the higher-level vocabulary, or about the possibility
of translating such sentences into another idiom. It is simply the claim that
every fact about currencies and the like is ultimately grounded in some perhaps
infinitely complex pattern of facts about other things. Now of course monetary
reductionism of this sort may not be true. That is a substantive question in the
philosophy of economics. The proposal is simply that our intuitive sense that
the US dollar is ontologically ‘derivative’ or ‘second-rate’ derives from our
strong suspicion that the dollar is unlikely to be a fundamental thing in the
sense defined above.
As formulated, the proposal entails that there are bona fide Things only if
there are fundamental facts. And it may be objected that we cannot assume this
a priori. For all we know, there may be infinite descending chains or trees in
which P is grounded in Q and R, which are in turn grounded in S, T, U and V,
and so on ad infinitum.
Q R
S T U V
As a crude model, we might imagine a world in which facts about atoms

are grounded in facts about quarks and electrons, which are in turn grounded in
facts about ‘hyperquarks’ and ‘hyperelectrons’, and so on forever. Of course,
the mere existence of some ungrounded trees would not entail that there are
no fundamental things, since there might still be fundamental facts elsewhere.
The worry arises only if every fact is grounded in further facts, in which case
our definitions entail that there are no fundamental things.
Is this an objectionable consequence of our proposal? The best reason
for thinking so runs as follows. Worlds in which every tree is ungrounded
are of two sorts. In some of these worlds, absolutely every object that is a
constituent of a fact at one level ultimately ‘disappears’ as we proceed down
the tree, chasing each fact into the facts that ground it. In that case our proposal
entails that while many things exist in the ordinary sense, nothing is ultimately
real. Rather everything has the same metaphysical status as the US dollar, and
that seems right. If old things dissolve and new things emerge as we boost
the magnification of our metaphysical microscope, then it seems right to say
that everything is on a par, metaphysically speaking, and that nothing has the
ultimate status that a fundamental thing would have if there were any.
But there is also the possibility that even though every fact is grounded in
further facts, certain things are nonetheless persistent, in the sense that when
x is a constituent of some fact [. . . x. . . ], it is also a constituent of every fact
in which [. . . x. . . ] is grounded. Intuitively (the objection runs) a persistent
thing is every bit as real as a fundamental thing would be, and a thing can
be persistent without being fundamental; so we should not say that a thing is
real if and only if it is a constituent of some fundamental fact. We should say
that a thing is real if and only if it is either fundamental or persistent.
In response, I suggest that the problem will not arise. Suppose that x is
a persistent thing in the intended sense, and consider the fact that x is self-
identical: [x = x]. It is totally unclear how a simple fact of this sort might be
grounded in more basic facts. Since this fact is fundamental, it follows that x is
a fundamental thing. And if this is right, we may retain our definitions, since
any persistent thing will also be a fundamental thing.
Recapitulation and outstanding problems
Philosophers who contemplate mathematics often find themselves pulled in
two directions. On the one hand, the spectacular success of mathematics on
its own terms and in the sciences, together with the rank obviousness of some
of its basic principles, inclines them to suppose that the claims of standard
mathematics must be true, and hence that mathematical objects of various sorts
must exist. On the other hand, the idea that mathematical objects are on a
metaphysical par with paradigmatic things—we’ve taken concrete particulars
as our examples, but we might also consider God and the angels if they
exist—strikes many philosophers as preposterous. We have been struggling
to formulate a position that does full justice to both of these tendencies, and
we have a proposal. Qualified realism about mathematical objects is the thesis
that while numbers and the like exist in the sense that sentences affirming their
existence are literally true, they are not fundamental, in the sense that every
fact in which the numbers figure is ultimately grounded in facts that do not
involve numbers as constituents.
We may note in passing that this account explains why certain idealist and
constructivist theories of mathematics have often been regarded as incompati-
ble with full-strength realism. The mathematical idealist (to use a single label
to cover a range of views) accepts that at least some of the core claims of
standard mathematics are literally true, but insists that the mathematical facts
are somehow grounded in our mathematical thought or activity. The crudest
versions of this approach identify the objects of mathematics with ideas in
the mind (but whose mind?). Subtler versions insist that the mathematical
truths are somehow grounded in our practices, insisting that it is only because
we have accepted a mathematical framework in which (e.g.) the ordinary
rules of arithmetical calculation are deemed valid that it is true to say that
235 + 657 = 892. Proponents of such views are often cagey about how it is
that we, by our practices, manage to conjure the objects of mathematics into
existence. It has always been obvious that we do not literally make the numbers
in the sense in which a builder makes a house. But if that analogy is a dead end,
what sense can we give to the idea that the numbers are somehow our creation?
I will not pursue the matter except to say that our framework provides a way
of stating this position. The general scheme for mathematical idealism is as
follows:
For every mathematical fact [A],
[A] obtains in virtue of. . . (something about the thoughts, activities or prac-
tices of conscious beings).
In fact this is a good general scheme for idealism and constructivism in
every area of philosophy. Berkeley’s idealism about external objects might be
understood as the claim that every fact involving tables and chairs is ultimately
grounded in facts about the mind of God. Mill’s secular idealism would substi-
tute facts about possible human sensations. Kantian Constructivism in ethics
might be the thesis that every fact about right and wrong is ultimately grounded
in facts about the judgements to which any rational agent is committed when
he deliberates about what to do. Social constructivism of the sort that used
to be fashionable in the humanities might be the view that facts about social
reality (and in the absurd limiting case, facts about reality tout court) are
ultimately grounded in facts about our epistemic practices for accepting and
rejecting claims about that reality. These views are sketchy in the extreme,
but they all have the right general form: facts of one sort—facts which do not
seem to concern human thought directly—are said to be grounded in facts
about human thought or practice. The characteristic objects of the higher-
level discourse—ordinary objects, social classes, moral obligations, etc.—all
disappear when we examine the underlying facts, to be replaced by items of
a different sort: representations, ideas, people and their practices, etc. These
radical forms of idealism may be mistaken. But they exert a perennial (and
mysterious) attraction for philosophers and philosophically minded scholars
in other disciplines, so it is presumably worth seeking a clear statement of
them, as well as an account of why they strike us as incompatible with an
unqualified realism of the domain in question. The present framework provides
the beginnings of such an account.15
Let us return now to the mathematical case. Our framework puts us in a
position to entertain reductionist theses of the form:
For every mathematical fact [A] in some area:
[A] obtains in virtue of (. . . some fact that does not involve mathematical
objects as constituents).
But having entertained such theses, how are we to assess them? As our
examples begin to show, there are many ways to associate the arithmetical
facts with facts in a more fundamental-seeming idiom—proof-theoretic facts,
set-theoretic facts, facts of pure modal logic, etc.—in which the objects of
arithmetic do not figure as constituents. These proposals are all materially
adequate: they pair truths with truths; and when properly constructed, they
preserve logical relations. Indeed, in many cases they respect an even more
stringent (though somewhat elusive) constraint, which we might call relevance.
As noted, any proof of an arithmetical statement A is readily converted by
trivial steps into a proof of its formalist or modal structuralist counterpart,
and vice versa. Speaking strictly there is a difference between the fact that
there are two primes between 15 and 20 and the fact that every possible omega
15 It is common to suppose that idealism and constructivism are opposed to full-

strength realism because they posit the mind-dependence of certain apparently mind-
independent facts. On the present account, this is half-right: these views are alternatives
to realism because they hold that certain facts depend on lower-level facts involving
objects of other kinds. That these lower-level facts involve minds is interesting, but
incidental. Formalism and structuralism in the philosophy of arithmetic are also alter-
natives to full-strength realism, and for the same reason, and yet in these cases the
underlying facts are in no way psychological.
sequence has two prime elements between its 16th and 21st elements, or the
fact that the sentence ‘There are two primes . . . ’ is provable in PAω . The facts
are different because they have different constituents. And yet the claims are
so close in mathematical significance that one is tempted to say that from a
mathematical point of view they represent the same fact. I have not said this,
because I do not know how to explain the notion of sameness of fact that the
claim involves. (My own account is much too fine-grained to support such
claims.) Still, one might say that any decent reduction of arithmetic must have
this feature: the facts to which the arithmetical facts are reduced must have
at least approximately the same mathematical significance as the arithmetical
originals. The worry is that even if we impose this constraint, there will be
many equally compelling reductionist proposals for arithmetic, and in general:
if there is one plausible reductionist account for some area of mathematics,
there will be many.16
This is a worry because it is natural to think these proposals cannot all be
correct. If a fact about the numbers obtains in virtue of some fact about the
provability of a sentence in PAω , it is implausible that it should also obtain in
virtue of some quite different fact about all omega sequences. Now it should
be stressed that over-determination of this sort is possible in principle. If we
have a disjunctive fact [p or q], where p and q are both true, then [p or q]
obtains in virtue of [p] and also in virtue of [q]. But the idea that a unified
domain of fact—the arithmetical facts—might be massively and systematically
overdetermined by facts from two or more rather different domains may seem
implausible. And yet if we reject this possibility, the reductionist must choose.
What basis could he possibly have for such a choice? It is hard to say.
One response to this predicament is a form of scepticism. If the competing
reductions are incompatible, and if we have no basis for choosing among them,
then surely, the only appropriate response is suspension of judgement. On this
view, questions about whether and how the mathematical facts are grounded in
more basic facts are fully meaningful but ultimately unanswerable. This is not
an absurd idea. Why on earth should we have the resources to answer every
abstruse metaphysical question that we can entertain? Still it would be good to
know whether we can resist this disappointing denouement.
Here is one possible way out. So far we have considered the matter from
the top down, as it were, beginning with the mathematical facts and asking
for more fundamental facts in which they might be grounded. This exercise
presupposes that when we consider an ordinary mathematical claim—e.g., the
claim there are two primes between 15 and 20—we have a single definite fact
in view, a fact whose basis we may then consider. But perhaps this is a mistake.
Suppose it turns out that for each putative basis for arithmetic—the formalist
basis, the modal structuralist basis, etc.—there is a distinct domain of numbers,
facts about which are constituted by the basic facts in question. On this view,
there is no such thing as the system of natural numbers. There are rather the
16 For a survey of reductionist proposals for analysis, see Burgess and Rosen (1997,
ch. B).
formalist numbers, facts about which are grounded in facts about provability in
PAω , the modal structuralist numbers, facts about which are grounded in facts
about all possible omega systems, and so on. Since the differences between
these systems of numbers make no mathematical difference, the language and
practice of mathematics will have had no occasion to distinguish them. If this
is right, then ordinary mathematical language will be rife with semantic inde-
terminacy. When I point to a river in the distance and say, ‘Let’s call that river
“Alph” ’, I introduce a meaningful word; but since I have not bothered to deter-
mine which of the many river-like objects in the vicinity I mean to pick out,
my word does not denote a single thing. Rather it ‘divides its reference’ over
a range of candidates—some a bit wider than others, some a bit longer, etc.
When I use the word in a sentence, I simply have not made up my mind about
which of these candidates I mean to denote. And in such cases it is misleading
to speak of the fact that Alph is in Xanadu. Since there are many river-like
objects in the vicinity, there are many such facts, the differences between which
simply do not matter for my purposes. Likewise, if there are many systems of
number-like objects distinguished only by the ways in which the facts about
them are grounded and not by any mathematically important features, it makes
no sense to speak of the fact that 235 + 657 = 892. There are rather many
equally qualified facts in the vicinity, each concerning numbers of some deter-
minate kind, each grounded in some determinate way in underlying facts.17
If this is the metaphysico-semantic situation, then it is no wonder that
we do not know how to answer questions about how the facts of arithmetic
are ultimately grounded. We do not know how to answer these questions
because they have a false presupposition, viz., that there is a unique system
of mathematical objects and determinate range of facts about them to which
the ordinary language of arithmetic manages to refer.
It is far from certain that this combination of metaphysical pluralism and
semantic indeterminacy is ultimately coherent. In order to address the issue
we would need a general theory of the conditions under which facts of one
kind give rise to or generate facts of another kind, and the project of producing
a general theory of this sort is genuinely daunting. But let us suppose for a
moment that the view is not only coherent but correct. We may then note that
even though questions about the numbers are to be rejected as badly posed, we
might still endorse a form of qualified realism about arithmetic. For it might
turn out that every candidate interpretation of the language of arithmetic takes
that language to describe a class of objects that disappear upon reduction. If
that is so, then it will still be true to say that arithmetical objects are not to be
17 We have a precedent for this within ordinary mathematics. In certain foundational

contexts it is important to distinguish the natural number 235 from the integer 235, the
real number 235.0, the complex number 235 + 0i, etc. But in many normal contexts
these differences do not matter. If someone invokes the fact that 235 + 657 = 892 in
such a context, his speech act misfires, in the sense that he fails to pick out a single fact
with his words, though the misfire may be harmless.
found among the fundamental things, and hence that the objects of arithmetic
are not ultimately real.
This brings us to the hardest question that arises within this framework.
Suppose that each putative reduction is associated with its own class of ‘num-
bers’, as above. This dissolves the dispute between partisans of competing
reductions. But there remains an intelligible dispute between the qualified
realist who says that every system of numbers is reducible in this way, and
the full-strength realist who says that at least one system of number-like items
exists at the fundamental level. In the framework we have been discussing,
these are both meaningful hypotheses. And yet nothing we have said indicates
a way of deciding which is right.
Some philosophers will be inclined to wield a principle of parsimony at
this point. They will say that in framing an account of fundamental reality,
we should aim to get by with as few things (or as few categories of things)
as possible. If ground-level numbers are not needed in the philosophy of
mathematics—if reducible numbers will always ‘do’—then ground-level num-
bers are dispensable and we theorists should reject them. To proceed in this
way is to assume a priori that fundamental reality is a sparsely populated realm
of clear skies and desert landscapes, and speaking personally, I see no basis
whatsoever for this assumption. But if parsimony cannot help us, we may find
ourselves in a quandary to which the only sensible response is suspense of
judgement. We may find ourselves pushed, in other words, to the view that
the question of qualified realism about arithmetic is fully meaningful and yet
unanswerable by any method we can imagine.
In some parts of philosophy, this sort of impasse is a sign that one’s
questions are badly posed. Is that so in this case? I don’t think so. I believe
that the grounding relation that figures in the formulation of this debate is
(or can be made to be) fully intelligible. The hard question of realism about
mathematics—i.e., the question whether some mathematical objects exist at
the fundamental level—is therefore clear. I confess that I have no idea how
one might go about answering it. But that is not to say that it is ultimately
unanswerable. Philosophers have not always distinguished ordinary ontologi-
cal questions—questions about what exists simpliciter—from deep ontological
questions about what exists at the fundamental level (if there is one). As a
result, we have no clear paradigms of inquiry into questions of the latter sort,
and so no clear sense of what it takes to establish a claim about what is
‘ultimately real’. Now that we have made the distinction, we can review the
record, so to speak, in order to ask whether we have examples in philosophy
or in physics or even in theology of arguments that bear specifically on claims
about fundamental reality. If we find plausible examples, we may be able to
draw explicit methodological morals from them. The way forward in the meta-
physics of mathematics may then consist in bringing these methodological
principles to bear on questions about the grounding of mathematical truths.
Is this project likely to bear fruit? At this point, it seems to me that we have no
clear basis for an answer.
Comment on Gideon Rosen’s
‘The reality of mathematical objects’
Timothy Gowers
At the beginning of Gideon Rosen’s contribution to this volume, he describes a

certain kind of philosophical position about mathematics. Ever since I myself
have had considered philosophical views about mathematics, they have been
of exactly that kind. Not being a professional philosopher, I have never tried to
work out a fully detailed defence of my views, but I have always been confident
that they are substantially correct.
If anyone is capable of shaking that kind of confidence, it is Rosen, who
has a remarkable ability to articulate and then closely analyse the philosophical
positions of others. He also has a way of putting forward views that I instinc-
tively react against, such as a realism about mathematical objects, but in a way
that makes them much less objectionable. For instance, if you say to Rosen
that you do not believe in a metaphysical realm where numbers are floating
around and enjoying various complicated relationships with each other, he
will tell you that he does not either. He may then ask you why you say that
there are infinitely many primes if you do not actually believe that that is the
case. After a short conversation of this kind, it becomes hard (for me at least)
to understand exactly what the difference is between non-realist views and
sophisticated realist ones of the kind that Rosen holds.
These issues have been greatly clarified for me now that I have read
Rosen’s contribution to this volume. He calls the kind of position he discusses
qualified realism: roughly speaking, qualified realism is the view that mathe-
matical objects do indeed exist, but are, as he puts it, ‘metaphysically second
rate’. His aim is to say what this could possibly mean. With this aim in mind,
he introduces a ‘grounding relation’ between facts: roughly speaking, if fact
A is grounded in fact B, then fact B is more fundamental than fact A and is
sufficient to explain it. (Rosen gives several different examples to illustrate
and clarify this notion.) The difference between a full realist and a qualified
realist, he then suggests, is that a realist believes that some mathematical facts
THE R E A L I T Y O F M AT H E M AT I C A L O B J E C T S : C O M M E N T 133
are fundamental (that is, not grounded in any further facts), whereas a qualified
realist believes that, while mathematical facts may be objectively true, they are
all ultimately grounded in non-mathematical facts. As he puts it:
When a philosopher tells you that numbers are not real in the metaphysical
sense (even though they exist in the mathematical sense) one thing he or
she might mean is this: that every fact in which a number figures ultimately
obtains in virtue of some collection of facts in which numbers do not figure
as constituents. (p. 125)
Rosen says that he does not himself endorse this view, but it resonates with
me in the way that he wants it to: I think that if I were to try to put forward a
full defence of my own qualified realism about numbers, I would indeed do so
by trying to show that facts about numbers are grounded in other facts. There
are several ways that one might go about this, just as there are several different
philosophical positions that fall under the banner of qualified realism. Rosen
discusses two of them in detail (neither of them the approach that I myself
would follow), and also discusses the coherence of the position in general.
One of the merits of Rosen’s proposal is that it replaces murky questions
about the reality of numbers and mathematical statements (murky because it
is often not clear what ‘real’ means) by the closely related but much clearer
question of whether all mathematical facts are grounded in other facts. To
answer this question is a large project, but it is also a clearly defined one that
philosophers, and perhaps even mathematicians, could get their teeth into.
10
Getting more out of mathematics
than what we put in
Mark Steiner
In The Emperor’s New Mind (Penrose, 1989), and more recently, in The
Road to Reality (Penrose, 2005), Professor Penrose has championed what he
calls Platonism, Plato’s mathematical world. Since Professor Penrose is to be
present at this Symposium, I thought it would be appropriate to discuss his
view, or, rather what I think his view should be, based on the thrust of his
published work.
What Penrose is after, as he explains in Road, is the objectivity of
mathematics1 —rather than the existence of ‘mathematical objects,’—whether,
as another speaker at this Symposium, Professor Rosen, puts it, mathematics
is a ‘subject with no object’. As Burgess and Rosen point out in their excellent
book (Burgess and Rosen, 1997), Platonism was hijacked by Quine, who
defined it as ‘quantifying over mathematical objects’, and it no longer means
what it used to mean. Book after book (most of them published by Oxford
University Press) appear with learned discourses about, for example, whether,
should ‘mathematical objects’ cease suddenly to exist, it would make any
observable difference.2
I think, therefore, it would be better to eliminate reference to Plato alto-
gether, and speak of concepts, rather than objects, and look to Descartes as the
1 A modern source for the idea of emphasizing ‘objectivity’ over ‘objects’ in the
philosophy of mathematics are the writings of Georg Kreisel.
2 This question, by the way, is meaningless according to Quine’s philosophy, since
according to his form of Platonism we cannot describe even the observable world
without ‘quantifying over’ mathematical objects. The indispensability of atoms is
different—we can describe the appearances without appeal to atoms, but we need
the atoms to explain those appearances. Quinian indispensability is, as my late and
lamented teacher Sidney Morgenbesser used to say, a form of Kantianism, which Quine
probably inherited from C. I. Lewis, and thus it is misleading to speak of the ‘Quine–
Putnam indispensability thesis’; Putnam uses indispensability in the latter sense.
real source of Penrose’s feeling that such concepts as that of the Mandelbrot
set are objective. After all, Penrose does not think that every concept is
objective (as did Plato), and neither did Descartes, as he puts the matter in
Meditation III:
. . . we must notice a point about ideas which do not contain true and
immutable natures but are merely ones which are invented and put together
by the intellect. Such ideas can always be split up by the same intellect, not
simply by an abstraction but by a clear and distinct intellectual operation,
so that any ideas which the intellect cannot split up in this way were clearly
not put together by the intellect. When, for example, I think of a winged horse
or an actually existing lion . . . I readily understand that I am also able to
think of a horse without wings, or a lion which does not exist . . . and so on;
hence these things do not have true and immutable natures. But if I think of a
triangle or a square . . . then whatever I apprehend as being contained in the
idea of a triangle—for example that its three angles are equal to two right
angles—I can with truth assert of the triangle . . .
What Descartes is saying here is that you get more out of mathematics
than what you put in; there is ‘latent information’ inherent in mathematical
ideas that is not contained in their verbal definition.3 This ‘latent information’
is part of the essence of these ideas, not put there by any mathematician. This
is, in my opinion, the real difference between mathematics and games. Take
the following position in a chess endgame (see Fig. 10.1).4
White needs to play 262 accurate moves to mate, assuming perfect play
on both sides. In many cases the moves ‘make no sense’, in the sense that one
cannot explain them without actually giving the entire tree of moves. Though
obviously this kind of thing is not ‘expected’ (the 50 rule move, which is
violated five times in the winning ‘line’, proves this), the surprise here is how
little is contained in the rules of chess, how little you reap for the investment.
Thus, the essence of chess is arbitrary, certainly not one of Descartes’ ‘true and
immutable essences’.
Compare this to one of Professor Penrose’s favorite examples, the ‘magi-
cal’ complex numbers. When they were introduced by the Italians as imaginary
solutions for equations over the reals, nobody could have predicted that they
would play the role of tying together the real numbers, as in the beautiful
equation
eπ i + 1 = 0
3 I think Descartes is saying more than just that some mathematical truths are
synthetic in the sense of Kant. For Kant, both 7 + 5 = 12 and the theorem about the
sum of the angles of a triangle are synthetic truths, but in the first case we don’t learn
anything of mathematical interest, since the sum had to be either one number or another.
The sum of the angles of a triangle is real information.
4 See http://www.chessbase.com/newsroom2.asp?id=239; thanks to Sylvain Cap-
pell for supplying this URL.
GETTING M O R E O U T O F M AT H E M AT I C S 137
A B C D E F G H
Fig. 10.1 White to play and mate in 262 moves.
which is a special case of Euler’s discovery
eiθ = cos(θ ) + i sin(θ )
Note that when imaginaries were introduced, the idea of raising a real
number to an imaginary power was unthinkable even to Cardano and Bombelli.
Yet once one thinks of the idea, it turns out that there is little or no choice in
how to proceed.
Another property not recognized by the originators of imaginary numbers
is the absolute value of a complex number, which emerges when we embed the
complex numbers in the Euclidean plane. Using this property, mathematicians
can explain facts about the real numbers; for example, why the real function
1
1+x2
, defined everywhere on the reals, is not equal to its expansion as a power
series 1 − x2 + x4 − · · · wherever |x| ≥ 1 (the complex numbers with absolute
value 1 form a circle around the origin, and the real number 1 and the imaginary
i lie on this circle. For i, the function, even when continued into the complex
plane, is not defined, since the denominator is zero. Standard theorems in
complex analysis do the rest).
One might say that the introduction of imaginaries is motivated in part by
computational convenience: Cardano uses imaginary numbers even in comput-
ing real roots of a cubic equation (which can appear in his famous formula).
Even so, the ‘latent information’ inherent in his idea goes far beyond calcula-
tional convenience. All in all, the following passage from The Emperor’s New
Mind could have been written by Descartes (had he been able to prophesy the
future of mathematics):
While at first it may seem that the introduction of such square roots of negative
numbers was just a device—a mathematical invention designed to achieve a
specific purpose—it later becomes clear that these objects are achieving far
more than that for which they were originally designed. As I mentioned above,
although the original purpose of introducing complex numbers was to enable
square roots to be taken with impunity, by introducing such numbers we find
that we get, as a bonus, the potentiality for taking any other kind of root
or for solving any algebraic equation whatever. Later we find many other
magical properties that these complex numbers possess, properties that we
had no inkling about at first. These properties are just there. They were not
put there by Cardano, nor by Bombelli, nor Wallis, nor Coates, nor Euler, nor
Wessel, nor Gauss, despite the undoubted farsightedness of these, and other,
great mathematicians; such magic was inherent in the very structure that
they gradually uncovered. When Cardano introduced his complex numbers,
he could have had no inkling of the many magical properties which were
to follow—properties which go under various names, such as the Cauchy
integral formula, the Riemann mapping theorem, the Lewy extension prop-
erty. These, and many other remarkable facts, are properties of the very
numbers, with no additional modifications whatever, that Cardano had first
encountered in about 1539.
Penrose (1989, pp. 96–97)
The Descartes/Penrose idea of ‘true and immutable essences’ in math-

ematical concepts, which I discussed in a different essay some years back
(Steiner, 2000), has nothing particularly to do with applications of mathematics
to nature. Nevertheless, and this is the thesis of the present essay, it quite often
turns out that the very latent mathematical information found in mathematical
concepts—even those introduced for ‘computational convenience’—provide
the most spectacular applications of mathematics in natural science. This
‘surplus value’ is particularly glaring in the application of ‘imaginary’ numbers
to ‘real’ nature. As in pure mathematics, the origin of these applications is in
calculational convenience, yet they end up in descriptive necessity. In what
follows, I will discuss some very well known facts (to those who have studied
Penrose’s works), so well known that we may forget just how remarkable they
are. From now on, we will be discussing what is called the ‘unreasonable
effectiveness’ of mathematics in natural science.
Euler’s discovery makes for a convenient way to represent rotations in the
plane without messy trigonometric formulas—namely, by a unit vector in the
complex plane. Composition of two rotations is given by multiplying two unit
vectors, giving a unit vector whose argument (angle) is the sum of the two
arguments.
The 19th century saw a number of attempts to generalize this convenience
to rotations in space. Euler had shown how to represent spatial rotations by
three angles (called today the ‘Euler angles’); it seemed reasonable to general-
ize the notion of complex numbers to three-dimensional vectors in a ‘complex
space’. These attempts failed, and Hamilton was forced to ascend to the fourth
dimension to get a vector space endowed with a multiplication, in analogy to
the complex numbers.5 The elements of this algebra he called ‘quaternions’
of the form a + bi + cj + dk where the multiplication of the alignment ele-
ments were governed by i2 = j2 = k2 = ijk = −1, an equation he carved into
a bridge in his excitement on finding these relations. In analogy to complex
numbers, the unit quaternions represent spatial rotations. Multiplying two unit
quaternions represents the composition of the two represented rotations, so we
have (using a later terminology) a homomorphism. The fact that quaternion
multiplication is not commutative is fine for this purpose, since the rotations
themselves are not commutative. There are, however, some puzzles: the special
cases of i, j, k represent rotations about the x, y, z axes, respectively—only
rotations of 180◦ , not 90◦ , otherwise the successive rotations represented by
i, j, k will not bring the axes back to their original alignment. (See Penrose,
2005, ch. 11, for elucidations.) This means that the three unit quaternions
−i, −j, −k all represent rotations of 540◦ (= 180◦ ) as well. This is true in
general—the negative of a unit quaternion represents the same rotation as the
quarternion itself. The homomorphism is not an isomorphism, but two to one.
Any continuous (sub)group r(θ ) of spatial rotations about a fixed axis can be
represented homomorphically by a continuous path q(θ ) of unit quaternions
such that q(θ + 2π ) = −q(θ ), while r(θ + 2π ) = r(θ ). Each rotation gets two
labels, or ‘parities’. The parity of a rotation is useless information, or so it
seemed.
Another method of extending the representation of plane rotations by
complex numbers of magnitude one to spatial rotations evolved at roughly
the same time. This was representation of rotations by 2 × 2 unitary matrices
of (ordinary) complex numbers. A unitary matrix M satisfies MM∗ = I (the
identity matrix), where M∗ is the matrix we get by transposing rows and
columns in M, and also replacing all four entries with their complex conjugates
(i.e., x + iy → x − iy). The determinant of a unitary matrix must be a complex
number with absolute value equal to unity, and restricting the matrices to those
with determinant +1, we arrive at what is today called SU(2), the ‘special
unitary’ group of 2 × 2 matrices. This method is the fruit of the efforts of
mathematicians like Cayley, Laguerre and others. But the earliest I could find
5 I am grateful to Sylvain Cappell and Stanley Ocken for improving the formulation
here. Professor Cappell brought to my attention a deep result of Frank Adams, the
English topologist, that only in dimension 1, 2, 4 and 8 does Euclidean space have a
vector space with a multiplication that satisfies: (a) there is a multiplicative identity
(left and right); (b) the product of two non-zero vectors is non-zero. For n = 8, the
multiplication is not associative.
explicitly the homomorphism between SU(2), and the rotation group is 1884,
when Felix Klein (1888, p. 34ff)6 lectured on the icosahedron group and two
years later used the homomorphism to calculate with rotations in his study of
the gyroscope (see Klein, 1922).
SU(2) turns out to be isomorphic to the unit quaternions, so there is also
a two-to-one homomorphism from SU(2) onto all the rotations. For example,
the SU(2) matrices
iθ
e 0
0 e−iθ
correspond to a rotation of 2θ around
the z-axis7 When θ = 0, the matrix
10
is the identity matrix I = corresponding to the null rotation; when
01
θ reaches π , the matrix is −I, and the rotation is a full rotation. It takes
two full rotations to bring the matrix back to I. Again, we have a kind of
gratuitous labelling of the rotations as positive or negative, rotations in the
interval (360◦ , 720◦ ) being with SU(2), and calculational convenience was the
rationale of introducing SU(2) in the first place.
What I find remarkable is that this very superfluous information turns out
to be the key to some of the fundamental features of our universe. For the
symmetry of the electron turned out precisely to be that of SU(2). When we
rotate the electron 360◦ its quantum mechanical description (wave function) is
multiplied by −1! To get the electron back to its initial state, one must give it
two full turns. Had mathematical physicists been much better calculators, they
might have missed one of the spectacular discoveries in the history of science.8
This is one of many examples in which human limitations, far from impeding
scientific progress, were responsible for it.9
6 Klein does not explictly state that spatial rotations can be represented by 3 × 3
real orthogonal matrices with determinant +1, so we cannot yet credit him with the
discovery that SU(2) is two-to-one homomorphic onto SO(3).
7 We set up another correspondence between self-adjoint matrices and 3 vectors in
space (see Sternberg (1994, §1.2) or Goldstein (1980, ch. 4) for details). Let A be a
self-adjoint matrix, and M a member of SU(2). Then conjugation by M, i.e., MAM∗ ,
yields a self-adjoint matrix corresponding to a rotated vector.
8 See also Hadamard (1954, pp. 128–189): ‘Most surprising—I should say
bewildering—facts of that kind are connected with the extraordinary marks of contem-
porary physics. In 1913, Elie Cartan, one of the first among French mathematicians,
thought of a remarkable class of analytic and geometric transformations in relation
to the theory of groups. No reason was seen, at that time, for special consideration
of those transformations except just their esthetic character. Then, some fifteen years
later, experiments revealed to physicists some extraordinary phenomena concerning
electrons, which they could only understand by the help of Cartan’s ideas of 1913.’
9 The limitations on observation made it impossible for Kepler to observe the
perturbations of one planet by another, and he published his ‘laws’ of planetary motion
as though they were not so perturbed. This enabled Newton to derive from these laws
(a) (b)
Fig. 10.2 (a) Icosahedron. (b) Carbon-60 molecule (buckyball).
The fact that the electron ‘lives’ in a two-dimensional complex vector space
and has SU(2) symmetry is detectable even on the macroscopic level. Consider
the carbon-60 molecule, which has the shape of a buckyball, or the truncated
icosahedron, which is made out of hexagons and pentagons like a soccer ball.
The buckyball would appear to have the symmetry I of an icosahedron: sixty
different rotations around its center leave the buckyball invariant.10 Yet if we
want to study the paramagnetic behaviour of the C-60 molecule in a magnetic
field that also has icosahedral symmetry we must ‘pull back’ to the members
of SU(2) that correspond to the members of I: the true symmetry group of the
buckyball, then, is the 120 member subgroup, G, of SU(2) that is two-to-one
homomorphic onto I (Chung et al., 1994, §9).
There is a story within a story here. Laguerre published in 1867 a ‘letter’ to
Hermite in which he defines matrix multiplication and discusses its properties
(Laguerre, 1867: 1898). In the course of this discussion, he introduces matrices
over the integers modulo p, presumably for studies in number theory, but in
any case, not for any physical application. If we consider 2 × 2 matrices, let
p = 5, and restrict ourselves to invertible matrices whose determinant is +1,
we get exactly 120 matrices forming a group, and this group turns out to be
isomorphic to G, a rather startling fact, which means that Laguerre’s number-
theoretic idea turns out to describe the electronic and magnetic properties of
the buckyball—surely we got more out than we put in (Chung et al., 1994, §2)!
Let us now return to the electron itself, which returns to itself only after two
turns. It might be thought that, since the electron ignores standard geometry, it
the inverse square law of gravitational force, and then study the resulting perturbations.
The ‘missing’ lines in the spectrum of hydrogen were actually there (though weak); had
they been detectable (then), a major discovery would have been missed.
10 See Chung and Sternberg (1993) for a nice exposition.
could have returned to itself after 3 or any other number of turns. Nevertheless,
there is something ‘natural’ about the number 2 which can be understood using
topology.
Using an example often cited, if you place the end of a belt in a book and
rotate the book a full turn, while holding the other end of the belt, you get a
twist in the belt. But it you turn the book two full turns, you can untwist the
belt by looping it under the book. Topologically this is expressed as follows: if
we make a closed loop of rotations, we cannot ‘shrink’ this loop into a ‘point’
(a point here meaning a constant curve of rotations, a curve that doesn’t rotate
anything). Only a double loop can be so shrunk. This is an indirect fact about
Euclidean space, which is revealed by studying the two-to-one homomorphism
between SU(2) and SO(3) (see Sternberg, 1994, §1.6).
Feynman has another “twist” (excuse the pun) on this idea (see Feynman
and Weinberg, 1987, pp. 56–59). He suggests tying a belt to two electrons,
and get them to switch places. The belt will be found to be twisted, showing
that topologically we have a single rotation, each electron having made what
is equivalent to half a rotation. In that case we should expect the wave function
describing the two electrons to change signs, which is the fundamental prop-
erty of fermions. If this argument is true, and it seems too good to be true, we
get the famous connection between ‘spin’ and ‘statistics’ using SU(2) and a
little topology—no relativity, no field theory. Even if this argument is incon-
clusive, there is a proof connecting SU(2) symmetry with fermionic statistics
using considerations that the inventors of SU(2) could not have dreamed of.
What is most remarkable, perhaps, is the further application of SU(2)
and unitary matrices in general. These further applications disconnect the
concept of unitary matrix more and more from their original application—the
representation of rotations.
I will cite two examples: the first is the application by Heisenberg to
nuclear physics in 1932 of SU(2) symmetry. It turned out that the neutron
and proton are two states of the same particle (called today the ‘nucleon’),
and that, mathematically, the two particles are analogous to the ‘spin up’ and
‘spin down’ states of the electron. The ‘rotation’ that would take the neutron
into the proton and then back to the neutron (a ‘rotation’ that would cause
the wave function of the particle to change sign) cannot be thought of as a
rotation in physical space; nevertheless, the SU(2) symmetry of the nucleon,
called ‘isotopic spin’, has empirical consequences. The physical basis for this
analogy, if there is one, is unknown even today.
When we move to higher-dimensional unitary matrices, the simple con-
nection with rotations breaks down. For example, SU(3) (the group of 3 × 3
unitary matrices with determinant +1), is not homomorphic onto any group
SO(n). Its connection to rotation is only via the analogy to SU(2). Despite
this, or perhaps because of this, SU(3) turns out to describe the three states of
a (type of) quark, which the nucleons are made of (since the nucleons have
integral charge, the quarks have fractional charge, a fact which impeded their
discovery). Rotations turned out to be the tip of a much bigger iceberg.
Another way to generalize the SU(2) two-to-one homomorphism onto the
rotation group was noted by Klein in his lectures on the top (1922, Lecture I).
If we drop the requirement of unitariness, and consider only 2 × 2 complex
matrices with determinant 1, we get the group of matrices today known as
SL(2,C). Klein shows (see p. 626 for a statement of this theorem) that there is
a two-to-one homomorphism between SL(2,C) and a group of transformations
on a two-dimensional manifold, and this group turns out to be isomorphic to
the proper Lorentz group L◦ (the group of Lorentz transformations that have
positive determinants and preserve the forward ‘light cone’), a homomorphism
which has itself great physical significance when we are studying relativistic
quantum theory.
I think that the applicability of ‘magical’ (Penrose’s phrase) complex num-
bers to rotations and its generalization to SU(2) and beyond is not unusual
in the history of mathematics, but it well illustrates the phenomenon that
mathematical concepts tend to have latent information, which can be used in
the development of mathematics. What is more, nature seems to make use of
these mathematical possibilities.
There are three elements coming together here: mathematics, nature, and
the human mind. Which of these three is responsible for the remarkable
richness of ‘mathematical essences’?
Acknowledgements
I would like to thank Shlomo Sternberg for crucial aid and advice. Thanks also
to Shmuel Elitzur, Sylvain Cappell, and Carl Posy for valuable discussions
over the course of many years. The research that went into this article was
supported by the Israel Science Foundation, Grant no. 251/06, and I am very
grateful for this support.
Comment on Mark Steiner’s ‘Getting
more out of mathematics than what
we put in’
Marcus du Sautoy
The idea of getting more out than you put in is one of the most attractive
features of my subject. I think this quotient is probably higher for mathematics
than for any of the other sciences, which is one reason I chose mathematics
over the mess of biology. Maybe it’s also because I have such a bad memory.
You only need a few axioms and everything else starts spilling out once you
start. Biology and chemistry always seemed to require you to memorize the
things Nature selected, which at times appear quite arbitrary.
The art of the mathematician is often to pick the rules of the game to
maximize this quotient. Although Steiner makes a distinction between games
and maths, I think there is the same satisfaction in seeing how the simple rules
of ‘Go’ lead to a game with such a rich array of plays, and the way three simple
axioms for the definition of a group can lead to the Monster simple group and
all the Lie groups that underpin physics.
Which brings me to the challenging point raised by Steiner’s article: the
extraordinary synergy between the abstract, beautiful world of mathematics
and the physical, messy world of physics, chemistry and biology. It is perhaps
bizarre that a perfect circle or right-angled triangle, the most fundamental
objects of mathematics, may not have any physical reality if quantum physics is
right about the world being quantized into discrete bits. And the infinities that
I bandy around with ease might have no concrete realization in a universe that
is potentially finite in nature. Nevertheless it is astounding how these objects
of the mind help us to predict the future behaviour of our messy universe. How
come the imaginary numbers we create to solve a cubic equation are the same
numbers that are crucial to describing the quantum world?
Perhaps it is an anthropomorphic answer as Steiner suggests in response
to my article. We make choices about the maths we like to celebrate. Where
did that maths and the excitement come from originally? From describing the
GETTING M O R E O U T O F M AT H E M AT I C S : C O M M E N T 145
physical world. The Egyptians wanted to know the volume of a pyramid. They
needed to know after all how many bricks to use. But to calculate the volume
they are led to the discovery of the power of cutting a shape into infinitely
many, infinitely thin pieces, which they can rearrange to make the problem
easier. An early form of integral calculus at work.
The process of cutting a real pyramid like this is clearly absurd on a practi-
cal level, yet a projection has been established from the world of mathematics
down onto our messy world. But because the world of mathematics began
it’s journey trying to describe and predict physical reality, perhaps it isn’t
so unexpected that the maths we generate in a purely abstract form, and for
its intrinsically internal fascination, nevertheless can often find itself being
projected back down to our messy universe generations after the journey was
kicked off.
A last point. Sometimes maths is very good at showing why you can’t get
any more out from what you put in. Real numbers led to complex numbers led
to quaternions and gave birth to octonions, but then mathematicians can prove
that you’re not going to get any more out of this. Similarly, the Lie groups E6 ,
E7 and E8 are such beautifully powerful structures, but the maths shows why it
stops there. There can’t be an E9 . Sometimes you get less out than you might
expect. But knowing that is sometimes as exciting as getting lots out of a small
investment. The exceptional Lie groups are special because of their unique
character. Still, it is amazing that E8 could be the model for the fundamental
particles that make up the fabric of reality. Nature certainly has good taste.
References
Archimedes. The Method. In Greek Mathematical Works. II: From Aristarchus

to Pappus, ed. J. Heiberg and trans. I. Thomas, Loeb Classical Library, 362.
Cambridge, MA: Harvard University Press (1993), pp. 221–223.
Arnauld, A. (1964). The Art of Thinking: Port Royal Logic, trans. J. Dickoff
and P. James. Indianapolis, IL: Bobbs-Merrill.
Benacerraf, P. (1965). What numbers could not be. Philosophical Review, 74.
Benacerraf, P. (1973). Mathematical truth. Journal of Philosophy, 70.
Bohm, D. and Hiley, B. J. (1993). The Undivided Universe. London: Rout-
ledge.
Bolzano, B. (1810). Contributions to a Better-Grounded Presentation of Math-
ematics. In The Mathematical Works of Bernard Bolzano, ed. and trans.
S. Russ. Oxford: Oxford University Press (2004).
Borges, J. L. (1941). The Library of Babel (La Biblioteca de Babel). In
Labyrinths. Harmondsworth: Penguin (1970).
Burgess, J. P. (1983). Why I am not a nominalist. Notre Dame Journal of
Formal Logic, 24: 1.
Burgess, J. and Rosen, G. (1997). A Subject with no Object: Strategies for
Nominalistic Interpretation of Mathematics. Oxford: Oxford University
Press.
Changeux, J.-P. and Connes, A. (1995). Conversations on Mind, Matter and
Mathematics, ed. and trans. M. B. DeBevoise. Princeton, NJ: Princeton
University Press.
Chung, F. and Sternberg, S. (1993). Mathematics and the buckyball. American
Scientist, 81: 56–71.
Chung, F., Kostant, B., and Sternberg, S. (1994). Groups and the buckyball. In
Lie theory and Geometry: In Honor of Bertram Kostant, ed. J.-L. Brylinski,
R. Brylinski, V. Guillemin and V. Kac. Boston: Birkhäuser.
Cicero, M. T. The Orations of Marcus Tullius Cicero. 4 vols. London: G. Bell
& Sons (1894–1903).
Coffa, A. (1986). From geometry to tolerance: sources of conventionalism in
nineteenth-century geometry. In From Quarks to Quasars: Philosophical
Problems of Modern Physics. University of Pittsburgh Series, 7. Pittsburgh,
PA: Pittsburgh University Press, 3–70.
148 R E F E R E N C E S
Cohen, P. J. (1966). Set Theory and the Continuum Hypothesis. New York:
W. A. Benjamin, Inc.
Courant, R. and Robbins, H. (1947). What is Mathematics? An Elementary
Approach to Ideas and Methods. Oxford: Oxford University Press (1981).
Curry, H. (1951). Outlines of a Formalist Philosophy of Mathematics. Amster-
dam: North-Holland.
Dedekind, R. (1888). Was sind und was sollen die Zahlen. In Gesammelte
Mathematische Werke, III. Braunschweig: Friedrich Vieweg und Sohn.
Detlefsen, M. (2005) Formalism. In The Oxford Handbook of Philosophy of
Mathematics and Logic, ed. S. Shapiro. Oxford: Oxford University Press,
pp. 236–317.
Dorr, C. (2008). There are no abstract objects. In Contemporary Debates in
Metaphysics, ed. T. Sider, J. Hawthorne and D. W. Zimmerman. Oxford:
Wiley-Blackwell.
du Sautoy, M. (2008). Finding Moonshine: A Mathematician’s Journey
Through Symmetry. London: Harper Perennial (2009).
Dummett, M. (1978). Truth and Other Enigmas. Cambridge, MA: Harvard
University Press.
Feynman, R. P. and Weinberg, S. (1987). Elementary Particles and the Laws
of Physics. Cambridge: Cambridge University Press.
Field, H. (1980). Science Without Numbers. Princeton, NJ: Princeton Univer-
sity Press.
Field, H. (1984). Is mathematical knowledge just logical knowledge?
Reprinted with a postscript in Realism, Mathematics, and Modality. Oxford:
Blackwell (1989), pp. 79–124.
Field, H. (1991). Metalogic and modality. Philosophical Studies, 62(1): 1–22.
Frege, G. (1884). The Foundations of Arithmetic: A Logico-Mathematical
Enquiry into the Concept of Number, trans. J. L. Austin, 2nd edn., New
York: Harper (1960); 2nd rev. edn., Evanston, IL: Northwestern University
Press (1968).
Frege, G. (1893). Grundgesetze der Arithmetik: begriffsschriftlich abgeleitet,
Vol. I. Hildesheim: G. Olms Verlag (1962).
Frege, G. (1903). Grundgesetze der Arithmetik: begriffsschriftlich abgeleitet,
Vol. II. Hildesheim: G. Olms Verlag (1962).
Gabriel, G. et al., eds. (1980). Gottlob Frege: Philosophical and Mathematical
Correspondence. Chicago: University of Chicago Press.
Gödel, K. (1947). What is Cantor’s continuum problem? Revised and
expanded version in Kurt Gödel: Collected Works, Vol. II. Oxford: Oxford
University Press (1990).
Gödel, K. (1951). Some basic theorems on the foundations of mathematics and
their implications. In Kurt Gödel: Collected Works, Vol. III. Oxford: Oxford
Goldstein, H. (1980). Classical Mechanics, 2nd edn. Reading, UK: Addison-
Wesley.
R E F E R E N C E S 149
Gregory, R. (1969). Eye and Brain: The Psychology of Seeing. New York:
McGraw Hill.
Hadamard, J. (1954). The Psychology of Invention in the Mathematical Field.
New York: Dover.
Hardy, G. H. (1940). A Mathematician’s Apology. Cambridge: Cambridge
Hartle, J. B. (2003). Gravity: An Introduction to Einstein’s General Relativity.
San Francisco: Addison-Wesley.
Hellman, G. (1989). Mathematics Without Numbers. Oxford: Oxford Univer-
sity Press.
Herschel, J. (1841). Review of Whewell’s History of the Inductive Sciences
and Philosophy of the Inductive Sciences. Quarterly Review, 68: 177–238.
Heyting, A. (1931). The intuitionistic foundations of mathematics. In Philoso-
phy of Mathematics, 2nd edn., ed. P. Benacerraf and H. Putnam. Cambridge:
Cambridge University Press (1983), pp. 52–61.
Hilbert, D. (1899). Die grundlagen der geometrie. In Festschrift zur Feier der
Enthullung des Gauss-Weber Denkmals in Göttingen. Leipzig: Teubner.
Horgan, T. (1994). Transvaluationism: a Dionysian approach to vagueness. The
Southern Journal of Philosophy, Supplement, 33: 97–126.
Hutton, C. (1795–1796). A Mathematical and Philosophical Dictionary. 2
vols. London: J. Johnson, and G. G. and J. Robinson. Reprinted, Hildesheim
and New York: G. Olms Verlag (1973), and in 4 vols., Bristol: Thoemmes
Press (2000).
Kant, I. (1781). Kritik der Reinen Vernunft, ed. R. Schmidt. Hamburg: Felix
Meiner Verlag (1990).
Kitcher, P. (1989). Explanatory unification and the causal structure of the
world. In Scientific Explanation, ed. P. Kitcher and W. Salmon. Minneapolis,
MI: University of Minnesota Press, pp. 410–505.
Klein, F. (1888). Lectures on the Ikosahedron and the Solution of Equations of
the Fifth Degree. London: Trübner & Co.
Klein, F. (1922). The mathematical theory of the top (1896/97). In Felix Klein
Gesmmelte Mathematische Abhandlungen, ed. R. Fricke and H. Vermeil.
Berlin: Springer.
Kreisel, G. (1967). Informal rigour and completeness proofs. In Problems in
the Philosophy of Mathematics, ed. I. Lakatos. Amsterdam: North-Holland,
pp. 138–186.
Laguerre, E. N. (1898). Sur le calcul des systemes linéaires. In Oeuvres de
Laguerre, ed. C. Hermite, H. Poincaré and R. Eugene. Paris: Gauthier-
Villars.
Lakatos, I. (1976). Proofs and Refutations, ed. J. Worrall and E. Zahar. Cam-
bridge: Cambridge University Press.
Leibniz, G. W. F. Die philosophischen Schriften von Gottfried Wilhelm Leibniz,
ed. C. J. Gerhardt and C. I. Gerhardt. Hildesheim: G. Olms Verlag (1978).
Leibniz, G. W. F. Discourse on Metaphysics and Other Essays, ed. D. Garber
and R. Ariew. Indianapolis, IL: Hackett (1989).
Leibniz, G. W. F. (1764). New Essays Concerning Human Understanding,
trans. A. G. Langley. Chicago: Open Court (1916).
Leibniz, G. W. F. (1683). Of Universal Analysis and Synthesis; or, of the Art
of Discovery and of Judgement. In Philosophical Writings [of] Leibniz, ed.
and trans. M. Morris and G. H. R. Parkinson. London: J. M. Dent and Sons
(1973), pp. 10–17.
Leibniz, G. W. F. Opera philosophica quae extant latina, gallica, germanica
omnia, ed. J. E. Erdmann. Aalen: Scientia (1959).
Leibniz, G. W. F. Opuscules et fragments inédits de Leibniz: extraits des
manuscrits de la bibliothèque royale de Hanover. Paris (1903).
Leslie, J. (1809). Elements of Geometry, Geometrical Analysis, and Plane
Trigonometry: with an appendix, notes and illustrations. Edinburgh: Brown
and Crombie.
Leslie, J. (1821). Geometrical Analysis, and Geometry of Curved Lines: being
volume second of a course of mathematics, and designed as an introduction
to the study of natural philosophy. Edinburgh: W. & C. Tait, and London:
Longman, Hurst, Rees, Orme, & Brown.
Lipton, P. (1991). Inference to the Best Explanation, 2nd edn. New York:
Routledge Publishing Company (2004).
Maddy, P. (2007). Second Philosophy: A Naturalistic Method. Oxford: Oxford
University Press.
Mates, B. (1986). The Philosophy of Leibniz: Metaphysics and Language.
Oxford: Oxford University Press.
Nelson, E. (1986). Predicative Arithmetic. Princeton, NJ: Princeton University
Press.
Penrose, R. (1989). The Emperor’s New Mind. Oxford: Oxford University
Press.
Penrose, R. (1994). Shadows of the Mind: An Approach to the Missing Science
of Consciousness. Oxford: Oxford University Press.
Penrose, R. (1997). On understanding understanding. International Studies in
the Philososophy of Science, 11: 7–20.
Penrose, R. (2004). The Road to Reality: A Complete Guide to the Laws of the
Universe. London: Jonathan Cape and New York: Alfred Knopf (2005).
Penrose, R. (2005). The Road to Reality: A Complete Guide to the Laws of the
Universe. New York: Alfred Knopf.
Penrose, R. (2011). Gödel, the mind, and the laws of physics. In Kurt Gödel
and the Foundations of Mathematics: Horizons of Truth, ed. M. Baaz,
C. H. Papadimitriou, D. S. Scott, H. Putnam and C. L. Harper, Jr. Cam-
bridge: Cambridge University Press, forthcoming.
Playfair, J. (1778). On the arithmetic of impossible quantities. Philosophical
Transactions of the Royal Society of London, 68: 318–343.
Polkinghorne, J. C. (1996). Beyond Science. Cambridge: Cambridge Univer-
sity Press.
Polkinghorne, J. C. (1998). Belief in God in an Age of Science. New Haven,
CT: Yale University Press.
R E F E R E N C E S 151
Polkinghorne, J. C. (2005). Exploring Reality. London: SPCK and New Haven,
CT: Yale University Press.
Proclus. A Commentary on the First Book of Euclid’s Elements, trans. G. R.
Morrow. Princeton, NJ: Princeton University Press (1970).
Putnam, H. (1967). Mathematics without foundations. Journal of Philosophy,
64.
Putnam, H. (1975). What is mathematical truth? In Mathematics, Matter and
Method, 2nd edn. Vol. 1 of Philosophical Papers. Cambridge: Cambridge
University Press (1979), pp. 60–78.
Quine, W. V. (1960). Word and Object. Cambridge, MA: The MIT Press.
Resnik, M. (1980). Frege and the Philosophy of Mathematics. Ithaca, NY:
Cornell University Press.
Resnik, M. (1997). Mathematics as a Science of Patterns. Oxford: Oxford
University Press.
Rosen, G. (1994). Objectivity and modern idealism. In Philosophy in Mind,
ed. J. O’Leary-Hawthorne and M. Michael. Dordrecht: Kluwer.
Rosen, G. (2006). Review of Jody Azzouni, deflating existential consequence.
Journal of Philosophy, 103: 6.
Rosen, G. (2010). Metaphysical dependence: reduction and grounding. In
Modality: Metaphysics, Logic and Epistemology, ed. B. Hale and A. Hoff-
mann. Oxford: Oxford University Press.
Rosen, G. and Burgess, J. P. (2005). Nominalism reconsidered. In Oxford
Handbook of Philosophy of Mathematics and Logic, ed. S. Shapiro. Oxford:
Oxford University Press.
Russell, B. (1905). On denoting. Mind, 14: 56.
Salmon, W. (1990). Four Decades of Scientific Explanation. Minneapolis, MI:
University of Minnesota Press, and Pittsburgh, PA: Pittsburgh University
Press (2006).
Schopenhauer, A. Arthur Schopenhauer’s Sämtliche Werke, Vol. 2. Munich:
R. Piper & Co. Verlag (1911).
Schopenhauer, A. (1859). The World as Will and Representation (Die Welt als
Wille und Vortstellung). New York: Dover Publications (1966).
Shapiro, S. (1997). Philosophy of Mathematics: Structure and Ontology.
Oxford: Oxford University Press.
Shapiro, S. (2000). The status of logic. In New Essays on the A Priori,
ed. P. Boghossian and C. Peacocke, Oxford: Oxford University Press,
pp. 333–366; reprinted (in part) as ‘Quine on Logic’, in Logica Yearbook
1999, ed. T. Childers, Prague: Czech Academy Publishing House, pp. 11–
21.
Shapiro, S. (2007). The objectivity of mathematics. Synthese, 156: 337–381.
Shapiro, S. (2007a). Vagueness in Context. Oxford: Oxford University Press.
Shapiro, S. (2009). We hold these truths to be self-evident: but what do we
mean by that? Review of Symbolic Logic, 2: 175–207.
Shapiro, S., ed. (2005). The Oxford Handbook of Philosophy of Mathematics
and Logic. Oxford: Oxford University Press.
Steiner, M. (1978). Mathematical explanation and scientific knowledge. Nous,
12: 17–28.
Steiner, M. (1980). Mathematical explanation. Philosophical Studies, 34:
135–152.
Steiner, M. (1998). The Applicability of Mathematics as a Philosophical Prob-
lem. Cambridge, MA: Harvard Uiniversity Press.
Steiner, M. (2000). Penrose and Platonism. In The Growth of Mathematical
Knowledge, ed. E. Grosholz and H. Breger. Dordrecht and Boston: Kluwer.
Sternberg, S. (1994). Group Theory and Physics. Cambridge: Cambridge
University Press.
Tait, P. G. (1866). Sir William Rowan Hamilton. North British Review, 14:
37–74.
Waismann, F. (1979). Ludwig Wittgenstein and the Vienna Circle. London:
Blackwell.
Waismann, F. (1982). Lectures on the Philosophy of Mathematics. Amsterdam:
Rodopi.
Weyl, H. (1921). Über die neue grundlagenkrise der mathematik. Mathematis-
che Zeitschrift, 10: 39–79.
Wittgenstein, L. (1953). Philosophical Investigations, trans. G. E. M.
Anscombe, 3rd edn. Oxford: Blackwell (2001).
Wittgenstein, L. (1956). Remarks on the Foundations of Mathematics, 3rd edn.
Oxford: Blackwell (1978); rev. edn., ed. G. H. von Wright, R. Rhees and G.
E. M. Anscombe, Cambridge, MA: The MIT Press (1967).
Woodward, J. (2009). Scientific explanation. Stanford Internet Encyclopedia
of Philosophy, http://plato.stanford.edu/entries/scientific-explanation.
Wright, C. (1992). Truth and Objectivity. Cambridge, MA: Harvard University
Press.
Index
Note: page numbers in italics refer to Figures.
632 symmetry group 18, 19 art

creation 5–6
A5 symmetry group 18 invention 4
abductive reasoning 65 Islamic 34
abstract artefacts 15 attributive reliability 78
abstract concepts, independent existence 5 axiomatization, constraints 64–5
acceptance 80–1 axioms 59
Adams, Frank 139 disputes over 110–11
aesthetic discourse, objectivity 109 forcedness 77, 93–4
aesthetics 26 Peano arithmetic 65–6
Islamic art 34
music 33 beauty of mathematics 20, 24, 26, 32–3, 37–8
role in mathematical creation 20, 22 Benacerraf, Paul, on reality of numbers 113
aisthēsis 73 Bohm, David, interpretation of quantum
Alhambra Palace, symmetrical designs 18, 19, physics 28
34 Bolzano, Bernard, on real definition 93
Ames, Adelbert, ‘Distorted Room’ 79 Borges, The Library of Babel 21
anagrams 4, 6 Braque, Georges, invention 4
analogy buckyball 141
argument by 29–30, 36 symmetry 141
between forcedness and sensory Burgess, John and Rosen, Gideon 135
perception 94 on theories of the world 99
anthropocentricity of mathematics 26
Applicability of Mathematics as a C*-Algebras, axioms 64
Philosophical Problem, The, calculations, consistency of results 30
M. Steiner 26 calculus, invention 6, 9, 14
applicability of mathematics 51, 56–7, Cantor, continuum hypothesis 44
68 Cappell, Sylvain 139
Archimedes, pre-demonstrative methods 82 Cardano, Girolamo, use of imaginary
argumentative stage of justification 82 numbers 137–8
Aristotle, on explanation 55 Cartan, Elie 140
arithmetic Carter, Howard 4
consistency of results 30 causal histories 52–3
formalization 117–19 causal model of explanation 50, 56, 57
implication of Gödel’s theorem 30 certificative stage of justification 82
mathematical unassailability 115 Changeaux, Jean-Pierre, view of mathematical
modal structuralism 119–21 reality 27, 28, 31
philosophical doubts 115–16 chess, comparison to mathematics 136
154 INDEX
Cicero, methods of systematic enquiry 81 in mathematics 19–20, 21
circle, Euclid’s definition 84, 86 musical 21
Classical Scheme of justification 82–4 unconscious 31
Coffa, Alberto, on theories of geometry 104 see also invention
cognitive command 100, 101–3 cricket, invention of rules 5
implications of vagueness 107–8 cryptography, application of number
as prerequisite for objectivity 112 theory 26
critique 109–11 culture, effect on mathematical discovery
qualifications 107 22
cognitive shortcoming 101
Cohen, P. J., technique of forcing 6, 9 ‘Death and the Maiden’ string quartet,
Columbus, Christopher 4, 6 Schubert 21
common sense 114–15 Dedekind, J.W.R., on real definition 93
complex number system see imaginary Dedekind–Peano axioms 64
numbers deductive reasoning 65
concepts definitial expansions 123
acquisition 89 definitions, value of 84–5
consistency 86–9, 92 demonstrative methods of investigation 82
Gödel’s view 74–6 dependency relations 57–8
Hermann Weyl’s view 90–1 ‘depth’, property of 33, 96
impossible 85 Descartes, René 136
Kant’s distinction from intuitions 74 determinate facts 123
real definition 84–6 Dirac, Paul, pursuit of mathematical
as a practical concern 86–9 beauty 33
as a theoretical concern 89–91 discovermental methods of investigation 82
Schopenhauer’s view 90 discovery 20
uninstantiated 91 cultural and historical context 22
conditioning 80–1 definition 4
Connes, Alain, view of mathematical distinction from observation 4
reality 27, 30 integration across mathematics 23
consciousness, physical basis 42–3 mathematical 6, 7–8
consistency of concepts 92 musical 21
role of real definitions 86–9 nature of 62
constraints philosophical perspective 13–15
in axiomatization 64–5 Plato’s Forms 62
in deductive reasoning 65 psychological aspects 11
role in sense of discovery 63–5 role of intuition 31
construction sense of 61, 70
du Sautoy’s symmetrical object 17–19 in deductive reasoning 65–6
of Monster group 6, 8, 14 Gödel’s view 76–7
constructivism 127, 127–8 illusory nature 67
continuum hypothesis 44 implications for mathematical
contradiction 85 reality 68–9
contrastive questions 53 role of constraints 63–5
conventionalism 68 discovery/invention question 3, 12, 27, 30–1,
Gödel’s view 75 92
convergence, as evidence for cognitive ancient views 81–4
command 103 implications of consistency 88
Conway’s game of Life 10 modern views 84–6
corpus delicti principle, jurisprudence 84 disjunctive facts 123
cosmological role 100 dispositions, effect on perception 79
creativity 5–6, 27 disputes, and cognitive command 101–8
in literature 21 ‘Distorted Room’ illusion 79
INDEX 155
divergent input, in characterization of fermions 142
cognitive command 105–6, 109 Feynman, Richard, on electron spin 142
fictionalization 86
Einstein, Albert, creative imagination 96 Field, Hartry, on logical possibility 67
electron, symmetry 140–2 forcedness
eliminative structuralism 121 detectability 92–4
elliptic curves, solutions to 17 felt objectivity as 71
epistemic constraint 100–1 as an indicator of reality 76, 77
epistemology, relationship to ontology 28 mathematical perceptions 78–80
eternal truths 22 sensory propositions 77–8
Euclid forcing, Cohen’s technique 6, 9
axioms for geometry 64 formal proof
definition of a circle 84, 86 definition of 103
Euclidean rescue 104 see also proof
of logic 105 formalism, arithmetic 117–18
Euler angles 139 formalist numbers 130
Euler’s equation 136–7 Forms 62, 74
surplus value 138–9 four-colour-map problem, proof 24
Euthyphro contrast 100 Frege, Gottlob
evolution of mathematical ability 31–2 on conditioning 81
‘exist’, use of term 114–15 on consistency 87–8, 92
existence on dependency relations 58
of mathematical objects 23, 26 Grundgesetze 71
see also mathematical reality fundamental facts 124, 125–6
as prerequisite for discovery 14–15
existential axioms, disputes over 110–11 Galileo
existential facts 123 gravitational acceleration thought
existential reliability 78 experiment 52
expectation, resistance to, as argument for Il Saggiatore 98
physical reality 95 games, comparison to mathematics 136, 144
explanation 55–6 Gelfrand–Naimark theorem 64
causal model 50 geometry, real definition
inference to the best explanation practical defence 86–7, 88
(IBE) 53–4 theoretical defence 89
interest-relativity 52–3 ‘Go’, comparison to mathematics 144
mathematical 56–7 Gödel
mathematical explanations of physical on concepts 74–5
phenomena 51, 56–7 on feelings of forcedness 71, 93–4
necessity model 50, 51–2 on forcedness of propositions 77–81
provision of understanding 49 phenomenological argument 76–7, 92
unification model 50–1 Gödel’s theorem 30, 66, 92
‘why-regress’ 49–50 as argument for mathematical reality 43–4
critique 46–7
facts 122–3 and completeness of PAω 118
fundamental 124 God’s eye view 99
familiarity, effect on mathematical good mathematics, quantification 24
perception 80 gravitational acceleration thought
‘felt objectivity’ 70–1 experiment 52
see also discovery: sense of Gregory, R. L., on Ames’ ‘Distorted Room’
Fermat, Pierre de 79
Last Theorem 22, 44 Griess, R., construction of Monster group 8
proof 24 grounding relation 122–4, 125–6, 131, 132–3
theorem of primes and squares 24 Grounding–Reduction Link 124
156 INDEX
group theory integration of mathematical discoveries
invention 8 23, 24
mathematical reality 95 interest-relativity of explanation 52–3
Grundgesetze, Gottlob Frege 71 intermediate value theorem 93
Grundlagen der Geometrie, David Hilbert 88 intuition 31, 73, 96, 97
intuitions, Kant’s distinction from concepts
Hadamard, Jacques 140 74
hallucinations, experience of forcedness 78 invention 4–5
Hamilton, W. R., discovery of abstract concepts 5
quaternions 63–4, 139 conclusions 12
hardness of the logical must, Wittgenstein 63 mathematical 6–7, 8, 9–10
Hardy, G. H. philosophical perspective 13–15
on mathematical reality 61 psychological aspects 11
A Mathematician’s Apology 19–20, 24, 26, see also discovery/invention question
30–1 Islamic art 34
Herchel, John, on consistency 87
Heyting, Arend, on intuition 97 jurisprudence, corpus delicti principle
Hilbert, David 84
Grundlagen der Geometrie 88 justification, Classical Scheme 82–4
proof-theory 92
historical context, effect on mathematical Kant, Immanuel 28
discovery 22 intuitions and concepts 74, 96
Horgan, Terrence, on independence 97–8 synthetic truths 136
human limitations, role in scientific Kant–Quine thesis, objectivity 99, 107
progress 140–1 Kepler, Johannes, laws of planetary
Hutton, C., on real definition 86 motion 140
hyperbolic geometry, discovery/invention 10 Kitcher, Philip 57
Klein, Felix 139
i knowing, distinction from understanding 49,
discovery/invention 6–7, 9–10 55
see also imaginary numbers knowledge, philosophical theory 116
idealism 99, 127–8 Kreisel, Georg 66, 135
ideas 74 Kronecker, Leopold 23
imaginary numbers
absolute value 137 Laguerre, E. N., on matrices 141
discovery/invention 6–7, 9–10 Lakatos, Imre, ‘monster-barring’ 107
Euler’s equation 136–7 language of the universe (Galileo) 98–9
motivation for introduction 137–8 latent information 136, 138, 143
surplus value ‘laws’, mathematical 57
quaternions 139 Leibniz, G. W. F.
unitary matrices 139–40 on definitions 84–5
impossible concepts 85 on impossible concepts 85–6
independent reality of mathematics 41, 42–4, invention of the calculus 6
97–8 Leslie, John 89
see also mathematical reality Library of Babel, The, Borges 21
indispensibility 135 Lie groups 144, 145
inductive reasoning 35–6 Life, game of 10
inference to the best explanation (IBE) 37, Lipton, Peter 55
53–4, 58–9, 62 on inference to the best explanation 37
Inference to the Best Explanation, Peter literary creativity 21, 23
Lipton 55 logic, objectivity 105–7
inferential error, in characterization of logical consequence 65–6
cognitive command 105–6 ‘felt objectivity’ 70–1
INDEX 157
logical possibility 66–7 for introduction of imaginary
logic-choice 106 numbers 137–8
Lorentz group 143 music, comparison to mathematics 23–4
musical appreciation 33
Mandelbrot set 6, 30 musical creativity 21, 23
materialism 27, 28 musical discoveries 21
unsatisfactoriness 29
mathematical objects 61–2 nature/nurture debate 24–5
Forms 62, 74 necessity model of explanation 50, 51–2, 56
minimal realism 114–16 neutron stars, PSR B1913+16 system 44
qualified realism 132–3 Newton, Isaac
modal structuralism example 119–21 invention of the calculus 6
reductionist examples 117–19 use of analogy 36
reality 113–14 noēsis 73
assessment of reductionism 128–31 noetic realm hypothesis 73
reductionist proposal 124–6 argument by analogy 29–30
reducibility 121–2 evolution as argument for 31–2
scientific confirmation 63 unreasonable effectiveness of mathematical
mathematical perception 80–1 beauty 32–4
distinction from sensory perception non-Euclidean geometry,
78–80 discovery/invention 7, 10, 20–1, 23
mathematical reality 20, 27, 61, 92 non-explanatory proofs 50–1
analogies with physical world 29–30 gravitational acceleration thought
and evolution of mathematical ability experiment 52
31–2 noumena 28
‘forcedness’ as an indicator 77–81 nucleons, SU(2) symmetry 142
Gödel’s theorem as argument for 43–4 number systems 129–30
critique 46–7 number theory, application to
Gödel’s view 76–7 cryptography 26
group theory 95 numbers
implications of mathematical Platonist view 124
discovery 68–9 reality 113–14, 115–16, 132–3
intuitive perception of 31 formalist view 119
metaphysics 28–9
theories as evidence 62 objectivity 97–8, 135–6
unreasonable effectiveness of mathematical cognitive command as prerequisite 112
beauty 32–4 critique 109–11
mathematical thinking, role of intuition 31 compromises 107
mathematical understanding 50 deductive reasoning 65–7
Mathematician’s Apology, A, G. H. ‘human’ influences on theorizing 99
Hardy 19–20, 24, 26, 30–1 Kant–Quine orientation 99
mental life, materialist view 29 of logic 105–7
metaphysical disputation 29 as a metaphysical concept 111
metaphysics 28–9 Wright’s account 100–1
objectivity 111 cognitive command 101–8
mind–brain relationship 28–9 observation, distinction from discovery 4
minimal realism 114–16 Ocken, Stanley 139
modal structuralism 119–21 omega sequences 120–1
modal structuralist numbers 130 ontology, relationship to epistemology 28
monetary reductionism 125 optical illusions
Monster group, discovery/invention 6, 8, 14 Ames’ ‘Distorted Room’ 79
‘monster-barring’ 107 experience of forcedness 78
motivation 20, 23–4, 144–5 ordered pairs, complex numbers as 9
158 INDEX
patterns of events, mathematical proof
explanation 51, 56 cognitive command 101–8
Peano arithmetic 117–19 creativity 23
Peano axioms 65–6 discovery/invention 7, 10–11, 65
Penrose, Roger 29 formal, definition of 103
on imaginary numbers 138 motivation 23–4
on objectivity 135–6 non-explanatory 50–1
perception, consistency 29–30 gravitational acceleration thought
perception-like experience of mathematical experiment 52
concepts 76 requirement of axioms 110
Permanence, Principle of 80 Protagoras 99
persistence 126 PSR B1913+16 neutron-star system 44
personal experience, materialist view 29 Putnam, H. 135
phenomena 28 pyramids, Egyptians’ calculation of
phenomenology of mathematics 61 volume 145
Gödel’s argument 76–7
physical behaviour, dependence on quadratic formula, discovery 7
mathematics 41–2, 44–5 qualified realism 114, 116–17, 125, 127,
physical explanation 132–3
causal model 50 modal structuralism example 119–21
necessity model 50 reductionist examples 117–19
physical phenomena, mathematical quantum physics
explanations 51, 56–7, 68 counterintuitive nature 30
physical realism 95–6 interpretation 28
physical world, analogies with reality 95
mathematics 30 quarks, SU(3) symmetry 142
physicalism 115 quaternions 139
physics W. R. Hamilton’s discovery 63–4
relationship to metaphysics 28 Quine, W. V. 135
search for beautiful equations 32–3 quintic, insolubility 7
1 -sentences 44
Picasso, Pablo Ramanujan, Srinivasa, intuitive perception
creation 6 31
invention 4 real definition 84–6, 92
planetary motion, Kepler’s laws 140 Bolzano’s views 93
Plato, Forms 62, 74 as a practical concern 86–9
Platonism 3, 23, 124, 135 as a theoretical concern 89–91
degrees of 43–4 realists 28
independent reality of mathematics 41, reality
42–4 dimensions of 27
physical behaviour, dependence on see also mathematical reality
mathematics 41–2, 44–5 recursively enumerable theories 118
Playfair, J. 86–7 reductio proofs 50
Poincaré, Henri, intuitive perception gravitational acceleration thought
31, 96 experiment 52
pre-demonstrative investigation 82 reductionism 117, 121–2, 124
primes assessment 128–31
discoveries 6 reflective equilibrium 106–7, 112
Fermat’s theorem 24 relevance 128
Riemann Hypothesis 22, 23 reliability, existential and attributive 78
Principle of Permanence 80 research, psychological aspects 11
problematic investigations 83 Resnik, Michael 104
Proclus, ‘ordering’ of Propositions 82–4 on wide reflective equilibrium 106
INDEX 159
richness, analogy between mathematics and surprise, analogy between mathematics and
physical world 30 physical world 30
Riemann Hypothesis 22, 23 symmetry
Rosen, Gideon, on dependency relations 57–8 designs of Alhambra Palace 18, 19, 34
rotations du Sautoy’s construction 17–19
representation by complex numbers synthetic truths 136
quaternions 139
unitary matrices 139–40 Taylor series expansion 68
symmetry of electron 140–2 theorematic investigations 83
Russell, Bertrand, on axioms of set theory 94 theories, discovery/invention 6, 14–15
theory development
Schopenhauer, on nature of concepts 90 axiomatizations 64–5
Schubert, ‘The Death and the Maiden’ string constraints 63–5
quartet 21 discovery of quaternions 63–4
science as evidence for mathematical reality 62
limited aims 28 Thompson, J. J. 4
richness of physical universe 30 thought, mind–brain relationship 28–9
scientific explanation 55–6 three ‘worlds’ 41, 42
scientific progress, role of human training, effect on mathematical perception
limitations 140–1 80
scientists, realism 28 transfinite numbers, discovery/invention 14
semantic consequence 66 Truth and Objectivity, Crispin Wright 100–1
semantic indeterminacy 130 cognitive command 101–8
sensory perception
analogy to forcedness 94 understanding
distinction from mathematical distinction from knowing 49, 55
perception 78–80 mathematical 50
sensory propositions, forcedness 77–8 unification model of explanation 50–1, 56, 57
set theory, axioms 77, 93–4 uninstantiated concepts 91
SL(2,C) matrices 143 unitary matrices 139–40
social constructivism 128 unreasonable effectiveness of mathematical
‘sporadic’ finite simple groups 30 beauty 33, 37–8
Steiner, Mark 57 utility of mathematics 20, 23, 26
strong reductionism 124
structuralism 119–21 vagueness, implications for cognitive
SU(2) symmetry 139–40 command 107–8
buckyball 141
electron 140, 142 Waismann, Friedrich, on Taylor series
nucleons 142 expansion 68
SU(3) symmetry 142 warfare, use of mathematics 26
supervenient facts 123 wave/particle duality of light 30
surplus value 138, 144 Weyl, Hermann, on nature of concepts 90–1
imaginary numbers why-questions, contrastive form 53
quaternions 139 ‘why-regress’ 49–50
properties of buckyball 141 wide reflective equilibrium 106–7
unitary matrices Wigner, Eugene, on mathematical beauty 33
139–40 Wittgenstein 63
SL(2,C) 143 on mathematical proof 67–8
SU(2) 141–2 Wright, Crispin, Truth and Objectivity 100–1
SU(3) 142 cognitive command 101–8

Meaning in Mathematics - John Polkinghorne

Uploaded by

Copyright:

Available Formats

Meaning in Mathematics - John Polkinghorne

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Meaning in Mathematics - John Polkinghorne

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

1 Is mathematics discovered or invented? 3

2 Exploring the mathematical library of Babel 17

4 Mathematics, the mind, and the physical world 41

6 Creation and discovery in mathematics 61

7 Discovery, invention and realism: Gödel and others

8 Mathematics and objectivity 97

9 The reality of mathematical objects 113

10 Getting more out of mathematics than what we put in 135

Editor: John Charlton Polkinghorne, KBE, FRS, the former president of

Michael Detlefsen is McMahon-Hank Professor of Philosophy at the Uni-

Is mathematics a highly sophisticated intellectual game in which the adepts

machines—that led to the building of many steam engines. Similarly, cricket

word or an invention-type word? If you type the phrases ‘complex numbers

A straightforward use of the word ‘invention’ in mathematics is to refer to

(a + bi)(c + di) = ac + bci + adi + bdi2 = (ad − bd) + (ad + bc)i

In a nearby possible world Timothy Gowers is not the distinguished

I’m a mathematician, not a philosopher. My job is to prove new theorems.

Fig. 2.1 The construction in my notebook of a new symmetrical object.

group was constructed that demonstrated the connection it might just be an

What I’d ‘constructed’ that evening in Bonn was an abstract symmetrical

Are mathematicians engaged in acts of discovery or are they merely construct-

A second argument appeals to the richness to be found in independent real-

Attacking the thorny issue of ‘creation vs. discovery’ in mathematics, John

Does mathematics have an independent reality? Or is it simply a product of

I am alas not a philosopher of mathematics, but I am interested in the nature

At the Castel Gandolfo Symposium, we had a most fruitful discussion of

Frege’s dependency relation is also asymmetric: no proposition grounds itself,

2 Georg Kreisel reportedly maintained that the question of realism in mathematics

It is not uncommon for those experienced in doing mathematics to see it as an

1 See Kant (1781: 1990, Bxxvi) for similar remarks.

Gödel’s phenomenological argument

3 Gödel expressly denied, however, that our perception-like experience of mathe-

‘Forcedness’ as an indicator of reality

There exists, unless I am mistaken, an entire world consisting of the totality of

7 Gödel, it will be recalled, maintained that ‘mathematics describes a non-sensual

8 The possibility of an effect of familiarity/training on belief-selection in mathemat-

. . . the essential logical and philosophical basis for operating in an extended

10 It was commonly believed as well that pre-demonstrative investigation ought to

11 That is, justiﬁcation capable of sustaining genuine scientiﬁc knowledge (scientia,

I: On a given ﬁnite straight line to construct an equilateral triangle.

Like other ancient geometers, Proclus’ conception of the distinction

Real deﬁnition as a practical concern

19 Leibniz’ example of a real deﬁnition was Euclid’s deﬁnition of a circle—namely,

20 By the ‘consistency’ of a concept is meant the consistency of the statements that

The practical identiﬁcation of demonstration of consistency with produc-

Similarly, in his correspondence with Hilbert concerning his (Hilbert’s)

Eventually, frustrated by Hilbert’s intransigency, he challenged him to

Frege’s reference to Hilbert’s Grundlagen der Geometrie was to his proof

21 In my view it is unclear how closely a model-construction proof like Hilbert’s

Real deﬁnition as a theoretical concern

23 There are of course serious questions concerning the similarity of model-

26 Russell and others, of course, expressly adopted this ‘regressive’ conception of

Michael Detlefsen gives us a careful discussion of the claim made by Kurt

A prominent, contemporary philosopher, Terrence Horgan (1994, p. 99)

He then adds, parenthetically that ‘sentences of pure mathematics are plausible

In other words, if epistemic constraint fails for a given discourse, then it

1 We will return to vagueness later.

2 I noted above that, according to Wright, cognitive command is an axis of objectiv-